Temperature variability, moisture and biotic interactions drive shrub species abundance along a coastal-inland gradient in arctic Greenland

- Secondary analysis workflow | Tasseled-cap Wetness Index as wetness predictor -

rm(list = ls())

Dependencies

Packages included:

• tidyverse - for convenient code flow, data wrangling and plotting
• rjags & R2jags - to link JAGS and R
• skimr - for data summary
• corrr - to output correlation matrices in dataframe format
• cowplot & patchwork - to combine plot panels
• Rmarkdown - to produce this neat little documentation

# load pacman package from the repository, if you do not already have it
if (!require('pacman')) install.packages('pacman', repos="https://cloud.r-project.org")
pacman::p_load(tidyverse, # set of packages for data manipulation, exploration and visualisation
               rjags,     # to link JAGS and R
               R2jags,    # to link JAGS and R
               skimr,     # for quick dataframe inspection
               corrr,     # output correlation matrices as data frame
               cowplot,   # combine plot panels
               patchwork, # -"-
               rmarkdown) # for R Markdown formatting

Functions used:

• effect size plot function

Colour scheme:

theme_darkgreen <- "#13944D"
theme_purple <- "#8757B3"

Overview

This project aims to disentangle drivers of abundance of different shrub species and functional groups in arctic tundra.

Analyses will integrate two levels of variation: plot group \(\times\) taxon level for climatic predictors (derived from downscaled CHELSA data), due to their coarser (90m) resolution, and plot \(\times\) taxon level for plot-level topographic and biotic interaction predictors.

The total dataset comprises 4968 observations for 66 variables (data compilation is documented in scripts/jonathan/nuuk_shrub_drivers_data_compilation.R):

env_cov_bio <- read.csv(file.path("..", "data", "processed", "nuuk_env_cover_plots.csv"), header = T)

Data preparation for JAGS models

a) selection of variables relevant for analysis

including predictors

• Information on site, plot, plot group, sampling location (lat, lon, altitude), sampling year
• downscaled CHELSA predictors, averaged over a 30-year period (*[…]_ts_30*)
• solar radiation index (SRI, following Keating et al. 2007), slope (erosion measure), Terrain Ruggedness Index (TRI, following Riley et al. 1999), Tasseled Cap Wetness Index (TCWI, Crist & Ciccone 1984), Topographic Wetness Index (TWI, see Conrad et al. 2015) based on FD8-Quinn and SAGA-internal flow algorithms, respectively
• cover of other shrub species (only used in species models)
• cover of graminoids
• competition pressure in the community (as summed abundance of taller-growing shrub species within a plot, averaged within plot groups)
• taxon (see below for levels)

and response variable

• cover (as relative no. hits per plot for each species/group)

env_cov_bio_sub <- env_cov_bio %>% 
  select(site_alt_plotgroup_id, plot, site, site_alt_id, year, long, lat, alt,  # plot info / metadata
  ends_with("_ts_30"),        # CHELSA predictors averaged over 30-year period prior to study year
  inclin_down, sri, tcws, tri, # environmental data
  ends_with("_cover"), compet,# biotic predictors: shrub & cover, relative competitiveness/acquisitiveness
  taxon, cover)               # taxon, cover response
head(env_cov_bio_sub)

Let’s check for correlation between the different moisture predictors and terrain ruggedness:

(cor_moist <- env_cov_bio %>% 
  select(twi_fd8, twi_saga, tcws, tri)) %>% 
  correlate(diagonal = 1)

Correlation method: 'pearson'
Missing treated using: 'pairwise.complete.obs'

\(\Rightarrow\) The two TWI are highly correlated with each other, and also rather highly correlated with TRI. All three are largely independent from TCWS. For this alternative pathway, we’ll go with TCWS to illustrate differences in model outcomes with different wetness predictors. Caution: as pointed out by reviewers, reflectance-based TCWS can be confounded by vegetation, though this might also make it more ‘evidential’ than the terrain-based TWI.

Predictors don’t always vary between plots within plotgroups - perhaps due to several falling into the same CHELSA grid cell.

Example: plots within site 1, altitude 20, plot group 1: plot P146 is slightly off and therefore has different climate variables than the other ones

Therefore plotgroup-scale (climatic) predictors will have to be averaged before using in models!

env_cov_bio %>% filter(taxon == "Betula nana" & site_alt_plotgroup_id == "1_20_1") %>% 
  ggplot(aes(x = long, y = lat)) + 
  geom_point() + 
  geom_text(aes(label = plot), hjust = 0.0001) + 
  xlim(c(-51.78675, -51.7863))

env_cov_bio %>% filter(taxon == "Betula nana" & site_alt_plotgroup_id == "1_20_1") %>% 
  select(site_alt_plotgroup_id, plot, tempjja_ts_30)

Check for correlation between predictors:

predictors_set <- env_cov_bio_sub %>% 
  select(ends_with("_ts_30"),   # CHELSA predictors averaged over 10-year period prior to study year
         inclin_down, sri, tri, tcws,      # environmental data
         shrub_cover, graminoid_cover, compet) %>%  # biotic predictors
  names()
# create basic correlation matrix
(cor_mat <- env_cov_bio_sub %>% 
  dplyr::select(predictors_set) %>% 
  correlate(diagonal = 1) %>% 
  # drop all values < .4 to increase readability
  mutate_if(is.numeric, ~ ifelse(abs(.) < .4, NA, .)))

Correlation method: 'pearson'
Missing treated using: 'pairwise.complete.obs'

\(\Rightarrow\) Maximum temperature is highly correlated with summer temperature, minimum temperature with continentality, slope (inclin_down) with solar radiation, and spring precipitation variables with summer precipitation. Let’s exclude them step by step and check the variance inflation factors (VIF) along the way. VIF values > 5 indicate collinearity issues.

# for the whole set of predictors
(vif_predictors_1 <- env_cov_bio_sub %>% 
  dplyr::select(predictors_set) %>% 
  usdm::vif()) #%>% View()

# => drop tempmax (correlated w/ tempjja)
(vif_predictors_2 <- env_cov_bio_sub %>% 
  dplyr::select(predictors_set,
                -tempmax_ts_30) %>% 
  usdm::vif()) #%>% View()

# => drop tempmin (correlated w/ tempcont)
(vif_predictors_3 <- env_cov_bio_sub %>% 
  dplyr::select(predictors_set,
                -tempmax_ts_30,
                -tempmin_ts_30) %>% 
  usdm::vif()) #%>% View()

# => drop precipjfmam (correlated w/ precipjja & tempcont)
(vif_predictors_4 <- env_cov_bio_sub %>% 
  dplyr::select(predictors_set,
                -tempmax_ts_30,
                -tempmin_ts_30,
                -contains("jfm")) %>% 
  usdm::vif()) #%>% View()

# => drop inclin_down (correlated w/ SRI)
(vif_predictors_5 <- env_cov_bio_sub %>% 
  dplyr::select(predictors_set,
                -tempmax_ts_30,
                -tempmin_ts_30,
                -inclin_down,
                -contains("jfm")) %>% 
  usdm::vif()) #%>% View()

# => drop precipmam (correlated w/ precipjja)
(vif_predictors_6 <- env_cov_bio_sub %>% 
  dplyr::select(predictors_set,
                -tempmax_ts_30,
                -tempmin_ts_30,
                -inclin_down,
                -contains("mam")) %>% 
  usdm::vif()) #%>% View()

All VIF are now < 3 -> OK! We can now exclude the dropped variables to obtain the final dataset. In addition, we rename the climate variables and delete the extension:

env_cov_bio_sub <- env_cov_bio_sub %>% 
  dplyr::select(-tempmax_ts_30,
                -tempmin_ts_30,
                -inclin_down,
                -contains("mam")) %>% 
  
  rename(tempjja = tempjja_ts_30,
         tempcont = tempcont_ts_30,
         precipjja = precipjja_ts_30) %>% 
  
  mutate(taxon = as.factor(taxon))
str(env_cov_bio_sub)

'data.frame':   4968 obs. of  28 variables:
 $ site_alt_plotgroup_id: chr  "1_20_1" "1_20_1" "1_20_1" "1_20_1" ...
 $ plot                 : chr  "P145" "P145" "P145" "P145" ...
 $ site                 : int  1 1 1 1 1 1 1 1 1 1 ...
 $ site_alt_id          : chr  "1_20" "1_20" "1_20" "1_20" ...
 $ year                 : int  2011 2011 2011 2011 2011 2011 2011 2011 2011 2011 ...
 $ long                 : num  -51.8 -51.8 -51.8 -51.8 -51.8 ...
 $ lat                  : num  64.2 64.2 64.2 64.2 64.2 ...
 $ alt                  : int  20 20 20 20 20 20 20 20 20 20 ...
 $ tempjja              : num  9.35 9.35 9.35 9.35 9.35 ...
 $ tempcont             : num  24.9 24.9 24.9 24.9 24.9 ...
 $ precipjja            : num  214 214 214 214 214 ...
 $ sri                  : num  0.933 0.933 0.933 0.933 0.933 ...
 $ tcws                 : num  -992 -992 -992 -992 -992 ...
 $ tri                  : num  0.28 0.28 0.28 0.28 0.28 ...
 $ graminoid_cover      : num  0.08 0.08 0.08 0.08 0.08 0.08 0.08 0.08 0.08 0.08 ...
 $ shrub_cover          : num  0.92 0.92 0.92 0.92 0.92 0.92 0.92 0.92 0.92 0.92 ...
 $ BetNan_shrub_cover   : num  0.92 0.92 0.92 0.92 0.92 0.92 0.92 0.92 0.92 0.92 ...
 $ CasTet_shrub_cover   : num  0.92 0.92 0.92 0.92 0.92 0.92 0.92 0.92 0.92 0.92 ...
 $ EmpNig_shrub_cover   : num  0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 ...
 $ PhyCae_shrub_cover   : num  0.92 0.92 0.92 0.92 0.92 0.92 0.92 0.92 0.92 0.92 ...
 $ RhoGro_shrub_cover   : num  0.92 0.92 0.92 0.92 0.92 0.92 0.92 0.92 0.92 0.92 ...
 $ RhoTom_shrub_cover   : num  0.92 0.92 0.92 0.92 0.92 0.92 0.92 0.92 0.92 0.92 ...
 $ SalArc_shrub_cover   : num  0.92 0.92 0.92 0.92 0.92 0.92 0.92 0.92 0.92 0.92 ...
 $ SalGla_shrub_cover   : num  0.72 0.72 0.72 0.72 0.72 0.72 0.72 0.72 0.72 0.72 ...
 $ VacUli_shrub_cover   : num  0.92 0.92 0.92 0.92 0.92 0.92 0.92 0.92 0.92 0.92 ...
 $ compet               : num  0.5988 -0.1508 -0.2 -0.0619 -0.0619 ...
 $ taxon                : Factor w/ 12 levels "All deciduous",..: 4 5 6 8 9 7 11 10 12 3 ...
 $ cover                : num  0 0 0.72 0 0 0 0.2 0 0 0.92 ...

b) adjustion of data structure

Data is ordered by site/altitude/plotgroup and taxon

env_cov_bio_sub <- env_cov_bio_sub[order(env_cov_bio_sub$site_alt_plotgroup_id, env_cov_bio_sub$taxon),]

As JAGS is only able to handle numeric input, all variables are assigned a numeric identifier:

env_cov_bio_sub$plotgroup.NUM <- as.numeric(factor(env_cov_bio_sub$site_alt_plotgroup_id,
                                                   levels = unique(env_cov_bio_sub$site_alt_plotgroup_id)))
env_cov_bio_sub$plot.NUM <- as.numeric(factor(env_cov_bio_sub$plot,
                                              levels = unique(env_cov_bio_sub$plot)))
env_cov_bio_sub$site_alt.NUM <- as.numeric(factor(env_cov_bio_sub$site_alt_id,
                                              levels = unique(env_cov_bio_sub$site_alt_id)))
env_cov_bio_sub$site.NUM <- as.numeric(factor(env_cov_bio_sub$site, levels = unique(env_cov_bio_sub$site)))
env_cov_bio_sub$taxon.NUM <- as.numeric(factor(env_cov_bio_sub$taxon, levels = unique(env_cov_bio_sub$taxon)))

Taxa were coded as follows:

data.frame(taxon = levels(env_cov_bio_sub$taxon),
           num = unique(env_cov_bio_sub$taxon.NUM))

Numeric predictors were scaled and centered:

num_pred <- env_cov_bio_sub %>% select(tempjja,
                                       tempcont,
                                       precipjja,
                                       sri, 
                                       tcws,
                                       tri, 
                                       ends_with("_cover"),
                                       matches("compet"))
for(i in 1:length(num_pred)){
  col <- colnames(num_pred[i])
  env_cov_bio_sub[paste0(col,"C")] <- as.numeric(scale(num_pred[i], scale = TRUE, center = TRUE))
}

To account for the range of the cover response (\(0 \leq cover \leq 1\)), the model needs a mixed structure incorporating a beta distribution (for all continuous values with \(0 < cover < 1\)) and a binomial distribution (for all discrete values of \(cover = \{0, 1\}\)). An additional variable cover_discrete was introduced to separate the dataset into discrete (= 1) and continuous (= 0) cover values:

env_cov_bio_sub$cover_discrete <- ifelse(env_cov_bio_sub$cover == 1 | env_cov_bio_sub$cover == 0, 1, 0)

JAGS models

The dataset was then ready to be split up into the species of interest. We create separate data subsets for all/discrete/continuous response variable values for each species:

# split dataframe by taxon
env_cov_bio_sub_spec.tot <- split(env_cov_bio_sub, env_cov_bio_sub$taxon)

# assign taxon name to list elements
# >> for total datasets
for (taxon_id in 1:nlevels(env_cov_bio_sub$taxon)){
  # extract 3-letter genus name string
  assign(paste0(str_extract(levels(env_cov_bio_sub$taxon)[taxon_id], 
                            "^\\w{3}"),
  # extract and capitalise 3-letter species name string
                str_to_title(str_remove(str_extract(levels(env_cov_bio_sub$taxon)[taxon_id],
                                                    "\\s\\w{3}"),
                                        "\\s")),
  # add extension
                ".tot"),
  # assign to respective list element
         env_cov_bio_sub_spec.tot[[taxon_id]])
}



# >> for discrete datasets
env_cov_bio_sub_spec.dis <- list()
for (taxon_id in 1:nlevels(env_cov_bio_sub$taxon)){
  # filter for discrete response values
  env_cov_bio_sub_spec.dis[[taxon_id]] <- filter(env_cov_bio_sub_spec.tot[[taxon_id]], cover_discrete == 1)
  # extract 3-letter genus name string
  assign(paste0(str_extract(levels(env_cov_bio_sub$taxon)[taxon_id], 
                            "^\\w{3}"),
  # extract and capitalise 3-letter species name string
                str_to_title(str_remove(str_extract(levels(env_cov_bio_sub$taxon)[taxon_id],
                                                    "\\s\\w{3}"),
                                        "\\s")),
  # add extension
                ".dis"),
  # assign to respective list element
         env_cov_bio_sub_spec.dis[[taxon_id]])
}

# >> for continuous datasets
env_cov_bio_sub_spec.cont <- list()
for (taxon_id in 1:nlevels(env_cov_bio_sub$taxon)){
  # filter for continuous response values
  env_cov_bio_sub_spec.cont[[taxon_id]] <- filter(env_cov_bio_sub_spec.tot[[taxon_id]], cover_discrete == 0)
  # extract 3-letter genus name string
  assign(paste0(str_extract(levels(env_cov_bio_sub$taxon)[taxon_id], 
                            "^\\w{3}"),
  # extract and capitalise 3-letter species name string
                str_to_title(str_remove(str_extract(levels(env_cov_bio_sub$taxon)[taxon_id],
                                                    "\\s\\w{3}"),
                                        "\\s")),
  # add extension
                ".cont"),
  # assign to respective list element
         env_cov_bio_sub_spec.cont[[taxon_id]])
}

# save all input data
# groups
save(list = c("AllShr.tot",
              "AllEve.tot",
              "AllDec.tot"),
     file = file.path("..", "data", "processed", "model_input_data_tcws", "shrub_gradient_group.datasets.Rdata"))

# species
# groups
save(list = c("BetNan.tot",
              "CasTet.tot",
              "EmpNig.tot",
              "PhyCae.tot",
              "RhoGro.tot",
              "RhoTom.tot",
              "SalArc.tot",
              "SalGla.tot",
              "VacUli.tot"),
     file = file.path("..", "data", "processed", "model_input_data_tcws", "shrub_gradient_species.datasets.Rdata"))

Groups

> assembling data for model input in lists

JAGS needs data input in list format, so all relevant variables are compiled into lists:

# Compile data into lists
# All shrubs ----
shrub_gradient_jags.AllShr.data <- list(
  
  # plot level predictors
  cov.tot = AllShr.tot$cover + 1,
  plotgroup.tot = AllShr.tot$plotgroup.NUM,
  sri.tot = AllShr.tot$sriC,
  tri.tot = AllShr.tot$triC,
  tcws.tot = AllShr.tot$tcwsC,
  gramin_cov.tot = AllShr.tot$graminoid_coverC,
  tempjja.tot.plot = AllShr.tot$tempjjaC,
  N_plots = length(unique(AllShr.tot$plot)),
  
  # plot group level predictors
  tempjja.tot = AllShr.tot %>% group_by(plotgroup.NUM) %>% summarise(tempjja.tot = mean(tempjjaC)) %>% pull(tempjja.tot), # one value per tXpg
  tempcont.tot = AllShr.tot %>% group_by(plotgroup.NUM) %>% summarise(tempcont.tot = mean(tempcontC)) %>% pull(tempcont.tot),
  precipjja.tot = AllShr.tot %>% group_by(plotgroup.NUM) %>% summarise(precipjja.tot = mean(precipjjaC)) %>% pull(precipjja.tot),
  N_plotgroups = length(unique(AllShr.tot$site_alt_plotgroup_id)),
  
  # subset of values for prediction, for each predictor...
  xhat_graminoid_cover = seq(from = min(AllShr.tot$graminoid_coverC), to = max(AllShr.tot$graminoid_coverC), length.out = 100),
  xhat_sri = seq(from = min(AllShr.tot$sriC), to = max(AllShr.tot$sriC), length.out = 100),
  xhat_tri = seq(from = min(AllShr.tot$triC), to = max(AllShr.tot$triC), length.out = 100),
  xhat_tcws = seq(from = min(AllShr.tot$tcwsC), to = max(AllShr.tot$tcwsC), length.out = 100),
  xhat_tempjja = seq(from = min(AllShr.tot$tempjjaC), to = max(AllShr.tot$tempjjaC), length.out = 100),
  xhat_precipjja = seq(from = min(AllShr.tot$precipjjaC), to = max(AllShr.tot$precipjjaC), length.out = 100),
  xhat_tempcont = seq(from = min(AllShr.tot$tempcontC), to = max(AllShr.tot$tempcontC), length.out = 100),
  Nxhat = 100,
  
  # ... and for predicting at high/low temperature levels
  xhat_tcws2 = as.numeric(c(quantile(AllShr.tot$tcwsC,0.05),quantile(AllShr.tot$tcwsC,0.95))), 
  Nxhat2 = 2
)

`summarise()` ungrouping output (override with `.groups` argument)
`summarise()` ungrouping output (override with `.groups` argument)
`summarise()` ungrouping output (override with `.groups` argument)

str(shrub_gradient_jags.AllShr.data)

List of 22
 $ cov.tot             : num [1:414] 1.52 1.84 1.6 1.72 1.52 1.2 1 1.16 1.64 1 ...
 $ plotgroup.tot       : num [1:414] 1 1 1 1 1 1 2 2 2 2 ...
 $ sri.tot             : num [1:414] 0.51 0.829 1.034 0.37 0.573 ...
 $ tri.tot             : num [1:414] -0.255 -0.314 -0.307 -0.274 -0.18 ...
 $ tcws.tot            : num [1:414] -0.4925 -0.0586 -0.0586 -0.0999 -0.6658 ...
 $ gramin_cov.tot      : num [1:414] -0.582 -0.389 -0.582 -0.582 -0.582 ...
 $ tempjja.tot.plot    : num [1:414] -0.321 -0.321 -0.321 -0.345 -0.345 ...
 $ N_plots             : int 414
 $ tempjja.tot         : num [1:69] -0.333 -0.286 -0.222 -0.396 -0.426 ...
 $ tempcont.tot        : num [1:69] -2.45 -2.52 -2.49 -2.41 -2.39 ...
 $ precipjja.tot       : num [1:69] 0.711 0.869 1.058 0.629 0.59 ...
 $ N_plotgroups        : int 69
 $ xhat_graminoid_cover: num [1:100] -0.582 -0.504 -0.426 -0.348 -0.27 ...
 $ xhat_sri            : num [1:100] -3.11 -3.06 -3.02 -2.98 -2.94 ...
 $ xhat_tri            : num [1:100] -1.72 -1.66 -1.6 -1.54 -1.48 ...
 $ xhat_tcws           : num [1:100] -3.08 -3.03 -2.97 -2.92 -2.87 ...
 $ xhat_tempjja        : num [1:100] -2.57 -2.52 -2.48 -2.44 -2.4 ...
 $ xhat_precipjja      : num [1:100] -1.66 -1.62 -1.58 -1.55 -1.51 ...
 $ xhat_tempcont       : num [1:100] -2.52 -2.49 -2.45 -2.42 -2.38 ...
 $ Nxhat               : num 100
 $ xhat_tcws2          : num [1:2] -1.62 1.62
 $ Nxhat2              : num 2

# All evergreen shrubs ----
shrub_gradient_jags.AllEve.data <- list(
  
  # plot level predictors
  cov.tot = AllEve.tot$cover + 1,
  plotgroup.tot = AllEve.tot$plotgroup.NUM,
  sri.tot = AllEve.tot$sriC,
  tri.tot = AllEve.tot$triC,
  tcws.tot = AllEve.tot$tcwsC,
  gramin_cov.tot = AllEve.tot$graminoid_coverC,
  tempjja.tot.plot = AllEve.tot$tempjjaC,
  N_plots = length(unique(AllEve.tot$plot)),
  
  # plot group level predictors
  tempjja.tot = AllEve.tot %>% group_by(plotgroup.NUM) %>% summarise(tempjja.tot = mean(tempjjaC)) %>% pull(tempjja.tot), # one value per tXpg
  tempcont.tot = AllEve.tot %>% group_by(plotgroup.NUM) %>% summarise(tempcont.tot = mean(tempcontC)) %>% pull(tempcont.tot),
  precipjja.tot = AllEve.tot %>% group_by(plotgroup.NUM) %>% summarise(precipjja.tot = mean(precipjjaC)) %>% pull(precipjja.tot),
  N_plotgroups = length(unique(AllEve.tot$site_alt_plotgroup_id)),
  
  # subset of values for prediction, for each predictor...
  xhat_graminoid_cover = seq(from = min(AllEve.tot$graminoid_coverC), to = max(AllEve.tot$graminoid_coverC), length.out = 100),
  xhat_sri = seq(from = min(AllEve.tot$sriC), to = max(AllEve.tot$sriC), length.out = 100),
  xhat_tri = seq(from = min(AllEve.tot$triC), to = max(AllEve.tot$triC), length.out = 100),
  xhat_tcws = seq(from = min(AllEve.tot$tcwsC), to = max(AllEve.tot$tcwsC), length.out = 100),
  xhat_tempjja = seq(from = min(AllEve.tot$tempjjaC), to = max(AllEve.tot$tempjjaC), length.out = 100),
  xhat_precipjja = seq(from = min(AllEve.tot$precipjjaC), to = max(AllEve.tot$precipjjaC), length.out = 100),
  xhat_tempcont = seq(from = min(AllEve.tot$tempcontC), to = max(AllEve.tot$tempcontC), length.out = 100),
  Nxhat = 100,
  
  # ... and for predicting at high/low temperature levels
  xhat_tcws2 = as.numeric(c(quantile(AllEve.tot$tcwsC,0.05),quantile(AllEve.tot$tcwsC,0.95))), 
  Nxhat2 = 2
)

`summarise()` ungrouping output (override with `.groups` argument)
`summarise()` ungrouping output (override with `.groups` argument)
`summarise()` ungrouping output (override with `.groups` argument)

str(shrub_gradient_jags.AllEve.data)

List of 22
 $ cov.tot             : num [1:414] 1.52 1.84 1.6 1.64 1.52 1.16 1 1.16 1.64 1 ...
 $ plotgroup.tot       : num [1:414] 1 1 1 1 1 1 2 2 2 2 ...
 $ sri.tot             : num [1:414] 0.51 0.829 1.034 0.37 0.573 ...
 $ tri.tot             : num [1:414] -0.255 -0.314 -0.307 -0.274 -0.18 ...
 $ tcws.tot            : num [1:414] -0.4925 -0.0586 -0.0586 -0.0999 -0.6658 ...
 $ gramin_cov.tot      : num [1:414] -0.582 -0.389 -0.582 -0.582 -0.582 ...
 $ tempjja.tot.plot    : num [1:414] -0.321 -0.321 -0.321 -0.345 -0.345 ...
 $ N_plots             : int 414
 $ tempjja.tot         : num [1:69] -0.333 -0.286 -0.222 -0.396 -0.426 ...
 $ tempcont.tot        : num [1:69] -2.45 -2.52 -2.49 -2.41 -2.39 ...
 $ precipjja.tot       : num [1:69] 0.711 0.869 1.058 0.629 0.59 ...
 $ N_plotgroups        : int 69
 $ xhat_graminoid_cover: num [1:100] -0.582 -0.504 -0.426 -0.348 -0.27 ...
 $ xhat_sri            : num [1:100] -3.11 -3.06 -3.02 -2.98 -2.94 ...
 $ xhat_tri            : num [1:100] -1.72 -1.66 -1.6 -1.54 -1.48 ...
 $ xhat_tcws           : num [1:100] -3.08 -3.03 -2.97 -2.92 -2.87 ...
 $ xhat_tempjja        : num [1:100] -2.57 -2.52 -2.48 -2.44 -2.4 ...
 $ xhat_precipjja      : num [1:100] -1.66 -1.62 -1.58 -1.55 -1.51 ...
 $ xhat_tempcont       : num [1:100] -2.52 -2.49 -2.45 -2.42 -2.38 ...
 $ Nxhat               : num 100
 $ xhat_tcws2          : num [1:2] -1.62 1.62
 $ Nxhat2              : num 2

# All decidious shrubs ----
shrub_gradient_jags.AllDec.data <- list(
  
  # plot level predictors
  cov.tot = AllDec.tot$cover + 1,
  plotgroup.tot = AllDec.tot$plotgroup.NUM,
  sri.tot = AllDec.tot$sriC,
  tri.tot = AllDec.tot$triC,
  tcws.tot = AllDec.tot$tcwsC,
  gramin_cov.tot = AllDec.tot$graminoid_coverC,
  tempjja.tot.plot = AllDec.tot$tempjjaC,
  N_plots = length(unique(AllDec.tot$plot)),
  
  # plot group level predictors
  tempjja.tot = AllDec.tot %>% group_by(plotgroup.NUM) %>% summarise(tempjja.tot = mean(tempjjaC)) %>% pull(tempjja.tot), # one value per tXpg
  tempcont.tot = AllDec.tot %>% group_by(plotgroup.NUM) %>% summarise(tempcont.tot = mean(tempcontC)) %>% pull(tempcont.tot),
  precipjja.tot = AllDec.tot %>% group_by(plotgroup.NUM) %>% summarise(precipjja.tot = mean(precipjjaC)) %>% pull(precipjja.tot),
  N_plotgroups = length(unique(AllDec.tot$site_alt_plotgroup_id)),
  
  # subset of values for prediction, for each predictor...
  xhat_graminoid_cover = seq(from = min(AllDec.tot$graminoid_coverC), to = max(AllDec.tot$graminoid_coverC), length.out = 100),
  xhat_sri = seq(from = min(AllDec.tot$sriC), to = max(AllDec.tot$sriC), length.out = 100),
  xhat_tri = seq(from = min(AllDec.tot$triC), to = max(AllDec.tot$triC), length.out = 100),
  xhat_tcws = seq(from = min(AllDec.tot$tcwsC), to = max(AllDec.tot$tcwsC), length.out = 100),
  xhat_tempjja = seq(from = min(AllDec.tot$tempjjaC), to = max(AllDec.tot$tempjjaC), length.out = 100),
  xhat_precipjja = seq(from = min(AllDec.tot$precipjjaC), to = max(AllDec.tot$precipjjaC), length.out = 100),
  xhat_tempcont = seq(from = min(AllDec.tot$tempcontC), to = max(AllDec.tot$tempcontC), length.out = 100),
  Nxhat = 100,
  
  # ... and for predicting at high/low temperature levels
  xhat_tcws2 = as.numeric(c(quantile(AllDec.tot$tcwsC,0.05),quantile(AllDec.tot$tcwsC,0.95))), 
  Nxhat2 = 2
)

`summarise()` ungrouping output (override with `.groups` argument)
`summarise()` ungrouping output (override with `.groups` argument)
`summarise()` ungrouping output (override with `.groups` argument)

str(shrub_gradient_jags.AllDec.data)

List of 22
 $ cov.tot             : num [1:414] 1 1 1 1.08 1 1.04 1 1 1 1 ...
 $ plotgroup.tot       : num [1:414] 1 1 1 1 1 1 2 2 2 2 ...
 $ sri.tot             : num [1:414] 0.51 0.829 1.034 0.37 0.573 ...
 $ tri.tot             : num [1:414] -0.255 -0.314 -0.307 -0.274 -0.18 ...
 $ tcws.tot            : num [1:414] -0.4925 -0.0586 -0.0586 -0.0999 -0.6658 ...
 $ gramin_cov.tot      : num [1:414] -0.582 -0.389 -0.582 -0.582 -0.582 ...
 $ tempjja.tot.plot    : num [1:414] -0.321 -0.321 -0.321 -0.345 -0.345 ...
 $ N_plots             : int 414
 $ tempjja.tot         : num [1:69] -0.333 -0.286 -0.222 -0.396 -0.426 ...
 $ tempcont.tot        : num [1:69] -2.45 -2.52 -2.49 -2.41 -2.39 ...
 $ precipjja.tot       : num [1:69] 0.711 0.869 1.058 0.629 0.59 ...
 $ N_plotgroups        : int 69
 $ xhat_graminoid_cover: num [1:100] -0.582 -0.504 -0.426 -0.348 -0.27 ...
 $ xhat_sri            : num [1:100] -3.11 -3.06 -3.02 -2.98 -2.94 ...
 $ xhat_tri            : num [1:100] -1.72 -1.66 -1.6 -1.54 -1.48 ...
 $ xhat_tcws           : num [1:100] -3.08 -3.03 -2.97 -2.92 -2.87 ...
 $ xhat_tempjja        : num [1:100] -2.57 -2.52 -2.48 -2.44 -2.4 ...
 $ xhat_precipjja      : num [1:100] -1.66 -1.62 -1.58 -1.55 -1.51 ...
 $ xhat_tempcont       : num [1:100] -2.52 -2.49 -2.45 -2.42 -2.38 ...
 $ Nxhat               : num 100
 $ xhat_tcws2          : num [1:2] -1.62 1.62
 $ Nxhat2              : num 2

# # save model input data
# shrub_gradient_jags.groupdata <- list

save(list = c("shrub_gradient_jags.AllShr.data",
              "shrub_gradient_jags.AllEve.data",
              "shrub_gradient_jags.AllDec.data"),
     file = file.path("..", "data", "processed", "model_input_data_tcws", "shrub_gradient_jags.groupdata.Rdata"))

> specifying model

write("
  
  model{
    
    # priors
      
      intercept ~ dnorm(0, 0.0001)
      
      b.gramin_cov ~ dnorm(0, 0.0001)
      b.sri ~ dnorm(0, 0.0001)
      b.tri ~ dnorm(0, 0.0001)
      b.tcws ~ dnorm(0, 0.0001)

      sigma.plot ~ dunif(0,100)
      tau.plot <- 1/(sigma.plot * sigma.plot)
      
      sigma.plotgroup ~ dunif(0,100)
      tau.plotgroup <- 1/(sigma.plotgroup * sigma.plotgroup)
      
      b.tempjja.x ~ dnorm(0, 0.001)
      b.tempjja.x2 ~ dnorm(0, 0.001)
      b.tempcont.x ~ dnorm(0, 0.001)
      b.tempcont.x2 ~ dnorm(0, 0.001)
      b.precipjja.x ~ dnorm(0, 0.001)
      b.precipjja.x2 ~ dnorm(0, 0.001)
      
      b.tempXtcws ~ dnorm(0, 0.001)
      b.tempXtcws2 ~ dnorm(0, 0.001)
      
      
    # plot level

      for (i in 1:N_plots){ 
        cov.tot[i] ~ dlnorm(mu.plot[i], tau.plot)
        log(mu.plot[i]) <- b_plotgroup[plotgroup.tot[i]] + # ~= random effect of plot group
                        b.gramin_cov * gramin_cov.tot[i] + 
                        b.tcws * tcws.tot[i] + 
                        b.tempXtcws * tempjja.tot.plot[i] * tcws.tot[i] +       # for interaction
                        b.tempXtcws2 * (tempjja.tot.plot[i]^2) * tcws.tot[i] +  # for interaction
                        b.sri * sri.tot[i] +
                        b.tri * tri.tot[i]
      }

    # plot group level
    
      for (k in 1:N_plotgroups){ # length of total plotgroups
        b_plotgroup[k] ~ dnorm(mu.plotgroup[k],tau.plotgroup)
        mu.plotgroup[k] <- intercept + 
                    
                    # plot group level predictors, linear and quadratic term
                    b.tempjja.x * tempjja.tot[k] + 
                    b.tempjja.x2 * (tempjja.tot[k]^2) + 
                    b.tempcont.x * tempcont.tot[k] + 
                    b.tempcont.x2 * (tempcont.tot[k]^2) +
                    b.precipjja.x * precipjja.tot[k] + 
                    b.precipjja.x2 * (precipjja.tot[k]^2) 
      }
      
      
      # add predicted values (derived parameters)
      for (m in 1:Nxhat){
        phat_graminoid_cover[m] <- intercept + b.gramin_cov * xhat_graminoid_cover[m]
        phat_sri[m] <- intercept + b.sri * xhat_sri[m]
        phat_tri[m] <- intercept + b.tri * xhat_tri[m]
        phat_tcws[m] <- intercept + b.tcws * xhat_tcws[m]
        phat_tempjja[m] <- intercept + b.tempjja.x * xhat_tempjja[m] + b.tempjja.x2 * (xhat_tempjja[m]^2)
        phat_tempcont[m] <- intercept + b.tempcont.x * xhat_tempcont[m] + b.tempcont.x2 * (xhat_tempcont[m]^2)
        phat_precipjja[m] <- intercept + b.precipjja.x * xhat_precipjja[m] + b.precipjja.x2 * (xhat_precipjja[m]^2)
      
        for (p in 1:Nxhat2){
          phat_tempXmoist[m,p] <- intercept +
                                      b.tempjja.x * xhat_tempjja[m] +
                                      b.tempjja.x2 * (xhat_tempjja[m]^2) +
                                      b.tcws * xhat_tcws2[p] +
                                      b.tempXtcws * xhat_tempjja[m] * xhat_tcws2[p] +
                                      b.tempXtcws2 * (xhat_tempjja[m]^2) * xhat_tcws2[p]
        }
      }

    
      }
  ", file.path("..", "models_tcws", "shrub_gradient.groups.jags"))

Specify the parameters to be monitored:

params_groups <- c("intercept",
                   "b.tempjja.x", "b.tempjja.x2",
                   "b.tempcont.x", "b.tempcont.x2",
                   "b.precipjja.x", "b.precipjja.x2",
                   "b.gramin_cov",
                   "b.sri",
                   "b.tri",
                   "b.tcws",
                   "b_plotgroup[1]","b_plotgroup[2]","b_plotgroup[3]","b_plotgroup[63]",
                   "sigma.plotgroup",
                   "phat_graminoid_cover", 
                   "phat_sri", "phat_tcws", "phat_tri", 
                   "phat_tempjja", "phat_tempcont", "phat_precipjja",
                   "phat_tempXmoist")

> run & evaluate model

All shrubs

# run model
model_out.shrub_gradient.AllShr <- jags(shrub_gradient_jags.AllShr.data,    # input data
                                        inits = NULL,                       # JAGS to create initial values
                                        params_groups,                      # parameters to be saved
                                        model.file = file.path("..", "models_tcws", "shrub_gradient.groups.jags"), 
                                        n.chains = 3,                       # no. Markov chains
                                        n.iter = 100000, n.burnin = 70000,  # no. iterations & burn-in fraction per chain
                                        n.thin = 2,                         # thinning rate
                                        DIC = FALSE,                        # do not compute deviance, pD, and DIC
                                        working.directory = NULL, 
                                        progress.bar = "text") 

# plot(model_out.shrub_gradient.AllShr) #check convergence, etc.

Extract coefficients and plot effect sizes:

# extract coefficients 
coeff.shrub_gradient.AllShr <- model_out.shrub_gradient.AllShr$BUGSoutput$summary %>% 
  as.data.frame %>% 
  select('mean','sd','2.5%','97.5%','Rhat') %>% 
# add identifying info to data frame
  rownames_to_column(var = "param")
  # mutate(param = as.vector(sapply(strsplit(rownames(coeff.shrub_gradient.AllShr),"[[]",fixed=FALSE), "[", 1))) #%>% print

# add 90% CIs
ci_90.AllShr <- data.frame(q5 = NA, q95 = NA, param = NA)
for (param in 1:(length(model_out.shrub_gradient.AllShr$BUGSoutput$sims.list)-4)){
  ci_90.AllShr[param,1:2] <- quantile(data.frame(model_out.shrub_gradient.AllShr$BUGSoutput$sims.list[param])[,1], probs = c(0.05, 0.95))
  ci_90.AllShr[param, 3] <- names(data.frame(model_out.shrub_gradient.AllShr$BUGSoutput$sims.list))[param]
}

# join to coefficients table
coeff.shrub_gradient.AllShr <- coeff.shrub_gradient.AllShr %>% 
  left_join(ci_90.AllShr, by = "param") %>% 
  # reorder and rename cols
  select(param, mean, sd, 
         l95 = "2.5%",
         l90 = q5,
         u90 = q95,
         u95 = "97.5%",
         Rhat) %>% print

save(coeff.shrub_gradient.AllShr, 
     file = file.path("..", "data", "processed", "model_outputs", "groups_tcws", "first_runs_converged", "model_output_AllShr.Rdata"))

(effect_size_plot.AllShr <- model_plot_sig_function(coeff.shrub_gradient.AllShr, title_string = "all shrubs", plot_width = 10.5))

• positive relationship with TC wetness, negative response to graminoid cover (sig.)
• slightly positive response to solar radiation (n.s.)

As none of the quadratic terms for climatic variables came out significant, re-run without them:

write("
  
  model{
    
    # priors
      
      intercept ~ dnorm(0, 0.0001)
      
      b.gramin_cov ~ dnorm(0, 0.0001)
      b.sri ~ dnorm(0, 0.0001)
      b.tri ~ dnorm(0, 0.0001)
      b.tcws ~ dnorm(0, 0.0001)

      sigma.plot ~ dunif(0,100)
      tau.plot <- 1/(sigma.plot * sigma.plot)
      
      sigma.plotgroup ~ dunif(0,100)
      tau.plotgroup <- 1/(sigma.plotgroup * sigma.plotgroup)
      
      b.tempjja.x ~ dnorm(0, 0.001)
      # b.tempjja.x2 ~ dnorm(0, 0.001)
      b.tempcont.x ~ dnorm(0, 0.001)
      # b.tempcont.x2 ~ dnorm(0, 0.001)
      b.precipjja.x ~ dnorm(0, 0.001)
      # b.precipjja.x2 ~ dnorm(0, 0.001)
      
      b.tempXtcws ~ dnorm(0, 0.001)
      # b.tempXtcws2 ~ dnorm(0, 0.001)
      
      
    # plot level

      for (i in 1:N_plots){ 
        cov.tot[i] ~ dlnorm(mu.plot[i], tau.plot)
        log(mu.plot[i]) <- b_plotgroup[plotgroup.tot[i]] + # ~= random effect of plot group
                        b.gramin_cov * gramin_cov.tot[i] + 
                        b.tcws * tcws.tot[i] + 
                        b.tempXtcws * tempjja.tot.plot[i] * tcws.tot[i] +       # for interaction
                        # b.tempXtcws2 * (tempjja.tot.plot[i]^2) * tcws.tot[i] +  # for interaction
                        b.sri * sri.tot[i] +
                        b.tri * tri.tot[i]
      }

    # plot group level
    
      for (k in 1:N_plotgroups){ # length of total plotgroups
        b_plotgroup[k] ~ dnorm(mu.plotgroup[k],tau.plotgroup)
        mu.plotgroup[k] <- intercept + 
                    
                    # plot group level predictors, linear and quadratic term
                    b.tempjja.x * tempjja.tot[k] + 
                    # b.tempjja.x2 * (tempjja.tot[k]^2) + 
                    b.tempcont.x * tempcont.tot[k] + 
                    # b.tempcont.x2 * (tempcont.tot[k]^2) +
                    b.precipjja.x * precipjja.tot[k] # + 
                    # b.precipjja.x2 * (precipjja.tot[k]^2) 
      }
      
      
      # add predicted values (derived parameters)
      for (m in 1:Nxhat){
        phat_graminoid_cover[m] <- intercept + b.gramin_cov * xhat_graminoid_cover[m]
        phat_sri[m] <- intercept + b.sri * xhat_sri[m]
        phat_tri[m] <- intercept + b.tri * xhat_tri[m]
        phat_tcws[m] <- intercept + b.tcws * xhat_tcws[m]
        phat_tempjja[m] <- intercept + b.tempjja.x * xhat_tempjja[m] # + b.tempjja.x2 * (xhat_tempjja[m]^2)
        phat_tempcont[m] <- intercept + b.tempcont.x * xhat_tempcont[m] # + b.tempcont.x2 * (xhat_tempcont[m]^2)
        phat_precipjja[m] <- intercept + b.precipjja.x * xhat_precipjja[m] # + b.precipjja.x2 * (xhat_precipjja[m]^2)
      
        for (p in 1:Nxhat2){
          phat_tempXmoist[m,p] <- intercept +
                                      b.tempjja.x * xhat_tempjja[m] +
                                      # b.tempjja.x2 * (xhat_tempjja[m]^2) +
                                      b.tcws * xhat_tcws2[p] +
                                      b.tempXtcws * xhat_tempjja[m] * xhat_tcws2[p] # +
                                      # b.tempXtcws2 * (xhat_tempjja[m]^2) * xhat_tcws2[p]
        }
      }

    
      }
  ", file.path("..", "models_tcws", "shrub_gradient.AllShr2.jags"))

# specify new set of parameters
params_AllShr2 <- c("intercept",
                   "b.tempjja.x", # "b.tempjja.x2",
                   "b.tempcont.x", # "b.tempcont.x2",
                   "b.precipjja.x", # "b.precipjja.x2",
                   "b.gramin_cov",
                   "b.sri",
                   "b.tri",
                   "b.tcws",
                   "b_plotgroup[1]","b_plotgroup[2]","b_plotgroup[3]","b_plotgroup[63]",
                   "sigma.plotgroup",
                   "phat_graminoid_cover", 
                   "phat_sri", "phat_tcws", "phat_tri", 
                   "phat_tempjja", "phat_tempcont", "phat_precipjja",
                   "phat_tempXmoist")

# run model
model_out.shrub_gradient.AllShr2 <- jags(shrub_gradient_jags.AllShr.data,    # input data
                                        inits = NULL,                       # JAGS to create initial values
                                        params_AllShr2,                      # parameters to be saved
                                        model.file = file.path("..", "models_tcws", "shrub_gradient.AllShr2.jags"), 
                                        n.chains = 3,                       # no. Markov chains
                                        n.iter = 100000, n.burnin = 70000,  # no. iterations & burn-in fraction per chain
                                        n.thin = 2,                         # thinning rate
                                        DIC = FALSE,                        # do not compute deviance, pD, and DIC
                                        working.directory = NULL, 
                                        progress.bar = "text") 

# plot(model_out.shrub_gradient.AllShr2) #check convergence, etc.

Extract coefficients and plot effect sizes:

# extract coefficients 
coeff.shrub_gradient.AllShr2 <- model_out.shrub_gradient.AllShr2$BUGSoutput$summary %>% 
  as.data.frame %>% 
  select('mean','sd','2.5%','97.5%','Rhat') %>% 
# add identifying info to data frame
  rownames_to_column(var = "param")
  # mutate(param = as.vector(sapply(strsplit(rownames(coeff.shrub_gradient.AllShr2),"[[]",fixed=FALSE), "[", 1))) #%>% print

# add 90% CIs
ci_90.AllShr2 <- data.frame(q5 = NA, q95 = NA, param = NA)
for (param in 1:(length(model_out.shrub_gradient.AllShr2$BUGSoutput$sims.list)-4)){
  ci_90.AllShr2[param,1:2] <- quantile(data.frame(model_out.shrub_gradient.AllShr2$BUGSoutput$sims.list[param])[,1], probs = c(0.05, 0.95))
  ci_90.AllShr2[param, 3] <- names(data.frame(model_out.shrub_gradient.AllShr2$BUGSoutput$sims.list))[param]
}

# join to coefficients table
coeff.shrub_gradient.AllShr2 <- coeff.shrub_gradient.AllShr2 %>% 
  left_join(ci_90.AllShr2, by = "param") %>% 
  # reorder and rename cols
  select(param, mean, sd, 
         l95 = "2.5%",
         l90 = q5,
         u90 = q95,
         u95 = "97.5%",
         Rhat) %>% print

save(coeff.shrub_gradient.AllShr2, file = file.path("..", "data", "processed", "model_outputs", "groups_tcws", "model_output_AllShr2.Rdata"))

(effect_size_plot.AllShr2 <- model_plot_sig_function(coeff.shrub_gradient.AllShr2, title_string = "all shrubs", plot_width = 7.5))

• positive response to TC wetness, negative response to graminoid cover (sig.)
• positive response to precipitation and solar radiation, negative to terrain heterogeneity (n.s.)

Evergreen shrubs

# run model
model_out.shrub_gradient.AllEve <- jags(shrub_gradient_jags.AllEve.data,    # input data
                                        inits = NULL,                       # JAGS to create initial values
                                        params_groups,                      # parameters to be saved
                                        model.file = file.path("..", "models_tcws", "shrub_gradient.groups.jags"), 
                                        n.chains = 3,                       # no. Markov chains
                                        n.iter = 100000, n.burnin = 70000,  # no. iterations & burn-in fraction per chain
                                        n.thin = 2,                         # thinning rate
                                        DIC = FALSE,                        # do not compute deviance, pD, and DIC
                                        working.directory = NULL, 
                                        progress.bar = "text") 

# plot(model_out.shrub_gradient.AllEve) #check convergence, etc.

Extract coefficients and plot effect sizes:

# extract coefficients 
coeff.shrub_gradient.AllEve <- model_out.shrub_gradient.AllEve$BUGSoutput$summary %>% 
  as.data.frame %>% 
  select('mean','sd','2.5%','97.5%','Rhat') %>% 
# add identifying info to data frame
  rownames_to_column(var = "param")
  # mutate(param = as.vector(sapply(strsplit(rownames(coeff.shrub_gradient.AllEve),"[[]",fixed=FALSE), "[", 1))) #%>% print

# add 90% CIs
ci_90.AllEve <- data.frame(q5 = NA, q95 = NA, param = NA)
for (param in 1:(length(model_out.shrub_gradient.AllEve$BUGSoutput$sims.list)-4)){
  ci_90.AllEve[param,1:2] <- quantile(data.frame(model_out.shrub_gradient.AllEve$BUGSoutput$sims.list[param])[,1], probs = c(0.05, 0.95))
  ci_90.AllEve[param, 3] <- names(data.frame(model_out.shrub_gradient.AllEve$BUGSoutput$sims.list))[param]
}

# join to coefficients table
coeff.shrub_gradient.AllEve <- coeff.shrub_gradient.AllEve %>% 
  left_join(ci_90.AllEve, by = "param") %>% 
  # reorder and rename cols
  select(param, mean, sd, 
         l95 = "2.5%",
         l90 = q5,
         u90 = q95,
         u95 = "97.5%",
         Rhat) %>% print

save(coeff.shrub_gradient.AllEve, file = file.path("..", "data", "processed", "model_outputs", "groups_tcws", "first_runs_converged", "model_output_AllEve.Rdata"))

(effect_size_plot.AllEve <- model_plot_sig_function(coeff.shrub_gradient.AllEve, title_string = "evergreen shrubs", plot_width = 10.5))

• unimodal relationship with temperature, positive with TC wetness, negative with graminoid cover (sig.)
• slightly positive response to solar radiation, slightly unimodal relationship with precipitation (n.s.)

write("
  
  model{
    
    # priors
      
      intercept ~ dnorm(0, 0.0001)
      
      b.gramin_cov ~ dnorm(0, 0.0001)
      b.sri ~ dnorm(0, 0.0001)
      b.tri ~ dnorm(0, 0.0001)
      b.tcws ~ dnorm(0, 0.0001)

      sigma.plot ~ dunif(0,100)
      tau.plot <- 1/(sigma.plot * sigma.plot)
      
      sigma.plotgroup ~ dunif(0,100)
      tau.plotgroup <- 1/(sigma.plotgroup * sigma.plotgroup)
      
      b.tempjja.x ~ dnorm(0, 0.001)
      b.tempjja.x2 ~ dnorm(0, 0.001)
      b.tempcont.x ~ dnorm(0, 0.001)
      # b.tempcont.x2 ~ dnorm(0, 0.001)
      b.precipjja.x ~ dnorm(0, 0.001)
      # b.precipjja.x2 ~ dnorm(0, 0.001)
      
      b.tempXtcws ~ dnorm(0, 0.001)
      b.tempXtcws2 ~ dnorm(0, 0.001)
      
      
    # plot level

      for (i in 1:N_plots){ 
        cov.tot[i] ~ dlnorm(mu.plot[i], tau.plot)
        log(mu.plot[i]) <- b_plotgroup[plotgroup.tot[i]] + # ~= random effect of plot group
                        b.gramin_cov * gramin_cov.tot[i] + 
                        b.tcws * tcws.tot[i] + 
                        b.tempXtcws * tempjja.tot.plot[i] * tcws.tot[i] +       # for interaction
                        b.tempXtcws2 * (tempjja.tot.plot[i]^2) * tcws.tot[i] +  # for interaction
                        b.sri * sri.tot[i] +
                        b.tri * tri.tot[i]
      }

    # plot group level
    
      for (k in 1:N_plotgroups){ # length of total plotgroups
        b_plotgroup[k] ~ dnorm(mu.plotgroup[k],tau.plotgroup)
        mu.plotgroup[k] <- intercept + 
                    
                    # plot group level predictors, linear and quadratic term
                    b.tempjja.x * tempjja.tot[k] + 
                    b.tempjja.x2 * (tempjja.tot[k]^2) + 
                    b.tempcont.x * tempcont.tot[k] + 
                    # b.tempcont.x2 * (tempcont.tot[k]^2) +
                    b.precipjja.x * precipjja.tot[k] # + 
                    # b.precipjja.x2 * (precipjja.tot[k]^2) 
      }
      
      
      # add predicted values (derived parameters)
      for (m in 1:Nxhat){
        phat_graminoid_cover[m] <- intercept + b.gramin_cov * xhat_graminoid_cover[m]
        phat_sri[m] <- intercept + b.sri * xhat_sri[m]
        phat_tri[m] <- intercept + b.tri * xhat_tri[m]
        phat_tcws[m] <- intercept + b.tcws * xhat_tcws[m]
        phat_tempjja[m] <- intercept + b.tempjja.x * xhat_tempjja[m] + b.tempjja.x2 * (xhat_tempjja[m]^2)
        phat_tempcont[m] <- intercept + b.tempcont.x * xhat_tempcont[m] # + b.tempcont.x2 * (xhat_tempcont[m]^2)
        phat_precipjja[m] <- intercept + b.precipjja.x * xhat_precipjja[m] # + b.precipjja.x2 * (xhat_precipjja[m]^2)
      
        for (p in 1:Nxhat2){
          phat_tempXmoist[m,p] <- intercept +
                                      b.tempjja.x * xhat_tempjja[m] +
                                      b.tempjja.x2 * (xhat_tempjja[m]^2) +
                                      b.tcws * xhat_tcws2[p] +
                                      b.tempXtcws * xhat_tempjja[m] * xhat_tcws2[p] +
                                      b.tempXtcws2 * (xhat_tempjja[m]^2) * xhat_tcws2[p]
        }
      }

    
      }
  ", file.path("..", "models_tcws", "shrub_gradient.AllEve2.jags"))

# specify new set of parameters
params_AllEve2 <- c("intercept",
                   "b.tempjja.x", "b.tempjja.x2",
                   "b.tempcont.x", # "b.tempcont.x2",
                   "b.precipjja.x", # "b.precipjja.x2",
                   "b.gramin_cov",
                   "b.sri",
                   "b.tri",
                   "b.tcws",
                   "b_plotgroup[1]","b_plotgroup[2]","b_plotgroup[3]","b_plotgroup[63]",
                   "sigma.plotgroup",
                   "phat_graminoid_cover", 
                   "phat_sri", "phat_tcws", "phat_tri", 
                   "phat_tempjja", "phat_tempcont", "phat_precipjja",
                   "phat_tempXmoist")

# run model
model_out.shrub_gradient.AllEve2 <- jags(shrub_gradient_jags.AllEve.data,    # input data
                                        inits = NULL,                       # JAGS to create initial values
                                        params_AllEve2,                      # parameters to be saved
                                        model.file = file.path("..", "models_tcws", "shrub_gradient.AllEve2.jags"), 
                                        n.chains = 3,                       # no. Markov chains
                                        n.iter = 100000, n.burnin = 70000,  # no. iterations & burn-in fraction per chain
                                        n.thin = 2,                         # thinning rate
                                        DIC = FALSE,                        # do not compute deviance, pD, and DIC
                                        working.directory = NULL, 
                                        progress.bar = "text") 

# plot(model_out.shrub_gradient.AllShr2) #check convergence, etc.

Extract coefficients and plot effect sizes:

# extract coefficients 
coeff.shrub_gradient.AllEve2 <- model_out.shrub_gradient.AllEve2$BUGSoutput$summary %>% 
  as.data.frame %>% 
  select('mean','sd','2.5%','97.5%','Rhat') %>% 
# add identifying info to data frame
  rownames_to_column(var = "param")
  # mutate(param = as.vector(sapply(strsplit(rownames(coeff.shrub_gradient.AllEve),"[[]",fixed=FALSE), "[", 1))) #%>% print

# add 90% CIs
ci_90.AllEve2 <- data.frame(q5 = NA, q95 = NA, param = NA)
for (param in 1:(length(model_out.shrub_gradient.AllEve2$BUGSoutput$sims.list)-4)){
  ci_90.AllEve2[param,1:2] <- quantile(data.frame(model_out.shrub_gradient.AllEve2$BUGSoutput$sims.list[param])[,1], probs = c(0.05, 0.95))
  ci_90.AllEve2[param, 3] <- names(data.frame(model_out.shrub_gradient.AllEve2$BUGSoutput$sims.list))[param]
}

# join to coefficients table
coeff.shrub_gradient.AllEve2 <- coeff.shrub_gradient.AllEve2 %>% 
  left_join(ci_90.AllEve2, by = "param") %>% 
  # reorder and rename cols
  select(param, mean, sd, 
         l95 = "2.5%",
         l90 = q5,
         u90 = q95,
         u95 = "97.5%",
         Rhat) %>% print

save(coeff.shrub_gradient.AllEve2, file = file.path("..", "data", "processed", "model_outputs", "groups_tcws", "model_output_AllEve2.Rdata"))

(effect_size_plot.AllEve2 <- model_plot_sig_function(coeff.shrub_gradient.AllEve2, title_string = "evergreen shrubs", plot_width = 8.5))

• unimodal response to temperature, negative response to graminoid cover and temp. variability, positive relationship with TCwetness (sig.)
• slightly positive response to solar radiation (n.s.)

Deciduous shrubs

# run model
model_out.shrub_gradient.AllDec <- jags(shrub_gradient_jags.AllDec.data,    # input data
                                        inits = NULL,                       # JAGS to create initial values
                                        params_groups,                      # parameters to be saved
                                        model.file = file.path("..", "models_tcws", "shrub_gradient.groups.jags"), 
                                        n.chains = 3,                       # no. Markov chains
                                        n.iter = 100000, n.burnin = 70000,  # no. iterations & burn-in fraction per chain
                                        n.thin = 2,                         # thinning rate
                                        DIC = FALSE,                        # do not compute deviance, pD, and DIC
                                        working.directory = NULL, 
                                        progress.bar = "text") 

# plot(model_out.shrub_gradient.AllDec) #check convergence, etc.

Extract coefficients and plot effect sizes:

# extract coefficients 
coeff.shrub_gradient.AllDec <- model_out.shrub_gradient.AllDec$BUGSoutput$summary %>% 
  as.data.frame %>% 
  select('mean','sd','2.5%','97.5%','Rhat') %>% 
# add identifying info to data frame
  rownames_to_column(var = "param")
  # mutate(param = as.vector(sapply(strsplit(rownames(coeff.shrub_gradient.AllDec),"[[]",fixed=FALSE), "[", 1))) #%>% print

# add 90% CIs
ci_90.AllDec <- data.frame(q5 = NA, q95 = NA, param = NA)
for (param in 1:(length(model_out.shrub_gradient.AllDec$BUGSoutput$sims.list)-4)){
  ci_90.AllDec[param,1:2] <- quantile(data.frame(model_out.shrub_gradient.AllDec$BUGSoutput$sims.list[param])[,1], probs = c(0.05, 0.95))
  ci_90.AllDec[param, 3] <- names(data.frame(model_out.shrub_gradient.AllDec$BUGSoutput$sims.list))[param]
}

# join to coefficients table
coeff.shrub_gradient.AllDec <- coeff.shrub_gradient.AllDec %>% 
  left_join(ci_90.AllDec, by = "param") %>% 
  # reorder and rename cols
  select(param, mean, sd, 
         l95 = "2.5%",
         l90 = q5,
         u90 = q95,
         u95 = "97.5%",
         Rhat) %>% print

save(coeff.shrub_gradient.AllDec, file = file.path("..", "data", "processed", "model_outputs", "groups_tcws", "first_runs_converged", "model_output_AllDec.Rdata"))

(effect_size_plot.AllDec <- model_plot_sig_function(coeff.shrub_gradient.AllDec, title_string = "deciduous shrubs", plot_width = 10.5))

• positive/unimodal response to temp. variability, negative response to graminoid cover (sig.)
• positive relationship with solar radiation, wetness; negative with terrain heterogeneity (m.s.)

write("
  
  model{
    
    # priors
      
      intercept ~ dnorm(0, 0.0001)
      
      b.gramin_cov ~ dnorm(0, 0.0001)
      b.sri ~ dnorm(0, 0.0001)
      b.tri ~ dnorm(0, 0.0001)
      b.tcws ~ dnorm(0, 0.0001)

      sigma.plot ~ dunif(0,100)
      tau.plot <- 1/(sigma.plot * sigma.plot)
      
      sigma.plotgroup ~ dunif(0,100)
      tau.plotgroup <- 1/(sigma.plotgroup * sigma.plotgroup)
      
      b.tempjja.x ~ dnorm(0, 0.001)
      # b.tempjja.x2 ~ dnorm(0, 0.001)
      b.tempcont.x ~ dnorm(0, 0.001)
      b.tempcont.x2 ~ dnorm(0, 0.001)
      b.precipjja.x ~ dnorm(0, 0.001)
      # b.precipjja.x2 ~ dnorm(0, 0.001)
      
      b.tempXtcws ~ dnorm(0, 0.001)
      b.tempXtcws2 ~ dnorm(0, 0.001)
      
      
    # plot level

      for (i in 1:N_plots){ 
        cov.tot[i] ~ dlnorm(mu.plot[i], tau.plot)
        log(mu.plot[i]) <- b_plotgroup[plotgroup.tot[i]] + # ~= random effect of plot group
                        b.gramin_cov * gramin_cov.tot[i] + 
                        b.tcws * tcws.tot[i] + 
                        b.tempXtcws * tempjja.tot.plot[i] * tcws.tot[i] +       # for interaction
                        # b.tempXtcws2 * (tempjja.tot.plot[i]^2) * tcws.tot[i] +  # for interaction
                        b.sri * sri.tot[i] +
                        b.tri * tri.tot[i]
      }

    # plot group level
    
      for (k in 1:N_plotgroups){ # length of total plotgroups
        b_plotgroup[k] ~ dnorm(mu.plotgroup[k],tau.plotgroup)
        mu.plotgroup[k] <- intercept + 
                    
                    # plot group level predictors, linear and quadratic term
                    b.tempjja.x * tempjja.tot[k] + 
                    # b.tempjja.x2 * (tempjja.tot[k]^2) + 
                    b.tempcont.x * tempcont.tot[k] + 
                    b.tempcont.x2 * (tempcont.tot[k]^2) +
                    b.precipjja.x * precipjja.tot[k] # + 
                    # b.precipjja.x2 * (precipjja.tot[k]^2) 
      }
      
      
      # add predicted values (derived parameters)
      for (m in 1:Nxhat){
        phat_graminoid_cover[m] <- intercept + b.gramin_cov * xhat_graminoid_cover[m]
        phat_sri[m] <- intercept + b.sri * xhat_sri[m]
        phat_tri[m] <- intercept + b.tri * xhat_tri[m]
        phat_tcws[m] <- intercept + b.tcws * xhat_tcws[m]
        phat_tempjja[m] <- intercept + b.tempjja.x * xhat_tempjja[m] # + b.tempjja.x2 * (xhat_tempjja[m]^2)
        phat_tempcont[m] <- intercept + b.tempcont.x * xhat_tempcont[m] + b.tempcont.x2 * (xhat_tempcont[m]^2)
        phat_precipjja[m] <- intercept + b.precipjja.x * xhat_precipjja[m] # + b.precipjja.x2 * (xhat_precipjja[m]^2)
      
        for (p in 1:Nxhat2){
          phat_tempXmoist[m,p] <- intercept +
                                      b.tempjja.x * xhat_tempjja[m] +
                                      # b.tempjja.x2 * (xhat_tempjja[m]^2) +
                                      b.tcws * xhat_tcws2[p] +
                                      b.tempXtcws * xhat_tempjja[m] * xhat_tcws2[p] # +
                                      # b.tempXtcws2 * (xhat_tempjja[m]^2) * xhat_tcws2[p]
        }
      }

    
      }
  ", file.path("..", "models_tcws", "shrub_gradient.AllDec2.jags"))

# specify new set of parameters
params_AllDec2 <- c("intercept",
                   "b.tempjja.x", # "b.tempjja.x2",
                   "b.tempcont.x", "b.tempcont.x2",
                   "b.precipjja.x", # "b.precipjja.x2",
                   "b.gramin_cov",
                   "b.sri",
                   "b.tri",
                   "b.tcws",
                   "b_plotgroup[1]","b_plotgroup[2]","b_plotgroup[3]","b_plotgroup[63]",
                   "sigma.plotgroup",
                   "phat_graminoid_cover", 
                   "phat_sri", "phat_tcws", "phat_tri", 
                   "phat_tempjja", "phat_tempcont", "phat_precipjja",
                   "phat_tempXmoist")

# run model
model_out.shrub_gradient.AllDec2 <- jags(shrub_gradient_jags.AllDec.data,    # input data
                                        inits = NULL,                       # JAGS to create initial values
                                        params_AllDec2,                      # parameters to be saved
                                        model.file = file.path("..", "models_tcws", "shrub_gradient.AllDec2.jags"), 
                                        n.chains = 3,                       # no. Markov chains
                                        n.iter = 100000, n.burnin = 70000,  # no. iterations & burn-in fraction per chain
                                        n.thin = 2,                         # thinning rate
                                        DIC = FALSE,                        # do not compute deviance, pD, and DIC
                                        working.directory = NULL, 
                                        progress.bar = "text") 

# plot(model_out.shrub_gradient.AllDec2) #check convergence, etc.

Extract coefficients and plot effect sizes:

# extract coefficients 
coeff.shrub_gradient.AllDec2 <- model_out.shrub_gradient.AllDec2$BUGSoutput$summary %>% 
  as.data.frame %>% 
  select('mean','sd','2.5%','97.5%','Rhat') %>% 
# add identifying info to data frame
  rownames_to_column(var = "param")
  # mutate(param = as.vector(sapply(strsplit(rownames(coeff.shrub_gradient.AllDec2),"[[]",fixed=FALSE), "[", 1))) #%>% print

# add 90% CIs
ci_90.AllDec2 <- data.frame(q5 = NA, q95 = NA, param = NA)
for (param in 1:(length(model_out.shrub_gradient.AllDec2$BUGSoutput$sims.list)-4)){
  ci_90.AllDec2[param,1:2] <- quantile(data.frame(model_out.shrub_gradient.AllDec2$BUGSoutput$sims.list[param])[,1], probs = c(0.05, 0.95))
  ci_90.AllDec2[param, 3] <- names(data.frame(model_out.shrub_gradient.AllDec2$BUGSoutput$sims.list))[param]
}

# join to coefficients table
coeff.shrub_gradient.AllDec2 <- coeff.shrub_gradient.AllDec2 %>% 
  left_join(ci_90.AllDec2, by = "param") %>% 
  # reorder and rename cols
  select(param, mean, sd, 
         l95 = "2.5%",
         l90 = q5,
         u90 = q95,
         u95 = "97.5%",
         Rhat) %>% print

save(coeff.shrub_gradient.AllDec2, file = file.path("..", "data", "processed", "model_outputs", "groups_tcws", "model_output_AllDec2.Rdata"))

(effect_size_plot.AllDec2 <- model_plot_marg_function(coeff.shrub_gradient.AllDec2, title_string = "deciduous shrubs", plot_width = 8.5))

• positive/unimodal relationship with temperature variability, negative relationship with graminoid cover, positive relationship with TC wetness and summer temperature (sig.)
• positive relationship with solar radiation, negative with terrain heterogeneity (m.s.)

Species

> species: assembling data for model input in lists

# BetNan ----
shrub_gradient_jags.BetNan.data <- list(
  
  # plot level predictors, for discrete...
  cov.dis = BetNan.dis$cover,
  plotgroup.dis = BetNan.dis$plotgroup.NUM,
  tempjja.dis = BetNan.dis$tempjjaC,
  sri.dis = BetNan.dis$sriC,
  tri.dis = BetNan.dis$triC,
  tcws.dis = BetNan.dis$tcwsC,
  shrub_cov.dis = BetNan.dis$BetNan_shrub_coverC,
  gramin_cov.dis = BetNan.dis$graminoid_coverC,
  compet.dis = BetNan.dis$competC,
  N_discrete = nrow(BetNan.dis),
  
  # ...and continuous part of the data
  cov.cont = BetNan.cont$cover,
  plotgroup.cont = BetNan.cont$plotgroup.NUM,
  tempjja.cont = BetNan.cont$tempjjaC,
  sri.cont = BetNan.cont$sriC,
  tri.cont = BetNan.cont$triC,
  tcws.cont = BetNan.cont$tcwsC,
  shrub_cov.cont = BetNan.cont$BetNan_shrub_coverC,
  gramin_cov.cont = BetNan.cont$graminoid_coverC,
  compet.cont = BetNan.cont$competC,
  N_cont = nrow(BetNan.cont),
  
  # plot group level predictors
  tempjja.tot = BetNan.tot %>% group_by(plotgroup.NUM) %>% summarise(tempjja.tot = mean(tempjjaC)) %>% pull(tempjja.tot), # one value per tXpg
  tempcont.tot = BetNan.tot %>% group_by(plotgroup.NUM) %>% summarise(tempcont.tot = mean(tempcontC)) %>% pull(tempcont.tot),
  precipjja.tot = BetNan.tot %>% group_by(plotgroup.NUM) %>% summarise(precipjja.tot = mean(precipjjaC)) %>% pull(precipjja.tot),
  N_plotgroups = length(unique(BetNan.tot$site_alt_plotgroup_id)),
  
  # subset of values for prediction, for each predictor...
  xhat_compet = seq(from = min(BetNan.tot$competC), to = max(BetNan.tot$competC), length.out = 100),
  xhat_shrub_cover = seq(from = min(BetNan.tot$BetNan_shrub_coverC), to = max(BetNan.tot$BetNan_shrub_coverC), length.out = 100),
  xhat_graminoid_cover = seq(from = min(BetNan.tot$graminoid_coverC), to = max(BetNan.tot$graminoid_coverC), length.out = 100),
  xhat_sri = seq(from = min(BetNan.tot$sriC), to = max(BetNan.tot$sriC), length.out = 100),
  xhat_tri = seq(from = min(BetNan.tot$triC), to = max(BetNan.tot$triC), length.out = 100),
  xhat_tcws = seq(from = min(BetNan.tot$tcwsC), to = max(BetNan.tot$tcwsC), length.out = 100),
  xhat_tempjja = seq(from = min(BetNan.tot$tempjjaC), to = max(BetNan.tot$tempjjaC), length.out = 100),
  xhat_precipjja = seq(from = min(BetNan.tot$precipjjaC), to = max(BetNan.tot$precipjjaC), length.out = 100),
  xhat_tempcont = seq(from = min(BetNan.tot$tempcontC), to = max(BetNan.tot$tempcontC), length.out = 100),
  Nxhat = 100,
  
  # ... and for predicting at high/low temperature levels
  xhat_tcws2 = as.numeric(c(quantile(BetNan.tot$tcwsC,0.05),quantile(BetNan.tot$tcwsC,0.95))), 
  Nxhat2 = 2
)

`summarise()` ungrouping output (override with `.groups` argument)
`summarise()` ungrouping output (override with `.groups` argument)
`summarise()` ungrouping output (override with `.groups` argument)

str(shrub_gradient_jags.BetNan.data)

List of 36
 $ cov.dis             : num [1:223] 0 0 0 0 0 0 0 0 0 0 ...
 $ plotgroup.dis       : num [1:223] 1 1 1 1 1 1 2 2 2 2 ...
 $ tempjja.dis         : num [1:223] -0.321 -0.321 -0.321 -0.345 -0.345 ...
 $ sri.dis             : num [1:223] 0.51 0.829 1.034 0.37 0.573 ...
 $ tri.dis             : num [1:223] -0.255 -0.314 -0.307 -0.274 -0.18 ...
 $ tcws.dis            : num [1:223] -0.4925 -0.0586 -0.0586 -0.0999 -0.6658 ...
 $ shrub_cov.dis       : num [1:223] 0.451 1.316 0.668 0.992 0.451 ...
 $ gramin_cov.dis      : num [1:223] -0.582 -0.389 -0.582 -0.582 -0.582 ...
 $ compet.dis          : num [1:223] 1.33 1.33 1.33 1.19 1.33 ...
 $ N_discrete          : int 223
 $ cov.cont            : num [1:191] 0.08 0.04 0.08 0.12 0.08 0.56 0.24 0.2 0.08 0.12 ...
 $ plotgroup.cont      : num [1:191] 5 5 6 10 10 11 11 11 11 12 ...
 $ tempjja.cont        : num [1:191] -0.42 -0.42 -0.322 1.009 1.009 ...
 $ sri.cont            : num [1:191] 0.823 -0.248 -0.663 -0.678 -0.192 ...
 $ tri.cont            : num [1:191] 0.4531 0.3713 -0.0771 -0.2698 0.1189 ...
 $ tcws.cont           : num [1:191] 0.176 -0.508 -0.83 -2.449 -2.548 ...
 $ shrub_cov.cont      : num [1:191] 2.397 1.1 -0.197 -0.629 -0.954 ...
 $ gramin_cov.cont     : num [1:191] 0.965 -0.582 -0.582 -0.195 -0.195 ...
 $ compet.cont         : num [1:191] 0.486 1.33 1.33 1.33 1.33 ...
 $ N_cont              : int 191
 $ tempjja.tot         : num [1:69] -0.333 -0.286 -0.222 -0.396 -0.426 ...
 $ tempcont.tot        : num [1:69] -2.45 -2.52 -2.49 -2.41 -2.39 ...
 $ precipjja.tot       : num [1:69] 0.711 0.869 1.058 0.629 0.59 ...
 $ N_plotgroups        : int 69
 $ xhat_compet         : num [1:100] -0.543 -0.524 -0.505 -0.487 -0.468 ...
 $ xhat_shrub_cover    : num [1:100] -0.954 -0.905 -0.855 -0.806 -0.757 ...
 $ xhat_graminoid_cover: num [1:100] -0.582 -0.504 -0.426 -0.348 -0.27 ...
 $ xhat_sri            : num [1:100] -3.11 -3.06 -3.02 -2.98 -2.94 ...
 $ xhat_tri            : num [1:100] -1.72 -1.66 -1.6 -1.54 -1.48 ...
 $ xhat_tcws           : num [1:100] -3.08 -3.03 -2.97 -2.92 -2.87 ...
 $ xhat_tempjja        : num [1:100] -2.57 -2.52 -2.48 -2.44 -2.4 ...
 $ xhat_precipjja      : num [1:100] -1.66 -1.62 -1.58 -1.55 -1.51 ...
 $ xhat_tempcont       : num [1:100] -2.52 -2.49 -2.45 -2.42 -2.38 ...
 $ Nxhat               : num 100
 $ xhat_tcws2          : num [1:2] -1.62 1.62
 $ Nxhat2              : num 2

# CasTet ----
shrub_gradient_jags.CasTet.data <- list(
  
  # plot level predictors, for discrete...
  cov.dis = CasTet.dis$cover,
  plotgroup.dis = CasTet.dis$plotgroup.NUM,
  tempjja.dis = CasTet.dis$tempjjaC,
  sri.dis = CasTet.dis$sriC,
  tri.dis = CasTet.dis$triC,
  tcws.dis = CasTet.dis$tcwsC,
  shrub_cov.dis = CasTet.dis$CasTet_shrub_coverC,
  gramin_cov.dis = CasTet.dis$graminoid_coverC,
  compet.dis = CasTet.dis$competC,
  N_discrete = nrow(CasTet.dis),
  
  # ...and continuous part of the data
  cov.cont = CasTet.cont$cover,
  plotgroup.cont = CasTet.cont$plotgroup.NUM,
  tempjja.cont = CasTet.cont$tempjjaC,
  sri.cont = CasTet.cont$sriC,
  tri.cont = CasTet.cont$triC,
  tcws.cont = CasTet.cont$tcwsC,
  shrub_cov.cont = CasTet.cont$CasTet_shrub_coverC,
  gramin_cov.cont = CasTet.cont$graminoid_coverC,
  compet.cont = CasTet.cont$competC,
  N_cont = nrow(CasTet.cont),
  
  # plot group level predictors
  tempjja.tot = CasTet.tot %>% group_by(plotgroup.NUM) %>% summarise(tempjja.tot = mean(tempjjaC)) %>% pull(tempjja.tot), # one value per tXpg
  tempcont.tot = CasTet.tot %>% group_by(plotgroup.NUM) %>% summarise(tempcont.tot = mean(tempcontC)) %>% pull(tempcont.tot),
  precipjja.tot = CasTet.tot %>% group_by(plotgroup.NUM) %>% summarise(precipjja.tot = mean(precipjjaC)) %>% pull(precipjja.tot),
  N_plotgroups = length(unique(CasTet.tot$site_alt_plotgroup_id)),
  
  # subset of values for prediction, for each predictor...
  xhat_compet = seq(from = min(CasTet.tot$competC), to = max(CasTet.tot$competC), length.out = 100),
  xhat_shrub_cover = seq(from = min(CasTet.tot$CasTet_shrub_coverC), to = max(CasTet.tot$CasTet_shrub_coverC), length.out = 100),
  xhat_graminoid_cover = seq(from = min(CasTet.tot$graminoid_coverC), to = max(CasTet.tot$graminoid_coverC), length.out = 100),
  xhat_sri = seq(from = min(CasTet.tot$sriC), to = max(CasTet.tot$sriC), length.out = 100),
  xhat_tri = seq(from = min(CasTet.tot$triC), to = max(CasTet.tot$triC), length.out = 100),
  xhat_tcws = seq(from = min(CasTet.tot$tcwsC), to = max(CasTet.tot$tcwsC), length.out = 100),
  xhat_tempjja = seq(from = min(CasTet.tot$tempjjaC), to = max(CasTet.tot$tempjjaC), length.out = 100),
  xhat_precipjja = seq(from = min(CasTet.tot$precipjjaC), to = max(CasTet.tot$precipjjaC), length.out = 100),
  xhat_tempcont = seq(from = min(CasTet.tot$tempcontC), to = max(CasTet.tot$tempcontC), length.out = 100),
  Nxhat = 100,
  
  # ... and for predicting at high/low temperature levels
  xhat_tcws2 = as.numeric(c(quantile(CasTet.tot$tempjjaC,0.05),quantile(CasTet.tot$tempjjaC,0.95))), 
  Nxhat2 = 2
)

`summarise()` ungrouping output (override with `.groups` argument)
`summarise()` ungrouping output (override with `.groups` argument)
`summarise()` ungrouping output (override with `.groups` argument)

str(shrub_gradient_jags.CasTet.data)

List of 36
 $ cov.dis             : num [1:398] 0 0 0 0 0 0 0 0 0 0 ...
 $ plotgroup.dis       : num [1:398] 1 1 1 1 1 1 2 2 2 2 ...
 $ tempjja.dis         : num [1:398] -0.321 -0.321 -0.321 -0.345 -0.345 ...
 $ sri.dis             : num [1:398] 0.51 0.829 1.034 0.37 0.573 ...
 $ tri.dis             : num [1:398] -0.255 -0.314 -0.307 -0.274 -0.18 ...
 $ tcws.dis            : num [1:398] -0.4925 -0.0586 -0.0586 -0.0999 -0.6658 ...
 $ shrub_cov.dis       : num [1:398] 0.101 0.911 0.304 0.607 0.101 ...
 $ gramin_cov.dis      : num [1:398] -0.582 -0.389 -0.582 -0.582 -0.582 ...
 $ compet.dis          : num [1:398] -0.266 -0.266 -0.266 -0.41 -0.266 ...
 $ N_discrete          : int 398
 $ cov.cont            : num [1:16] 0.04 0.04 0.08 0.04 0.08 0.72 0.08 0.04 0.16 0.04 ...
 $ plotgroup.cont      : num [1:16] 30 30 30 38 43 49 49 49 49 49 ...
 $ tempjja.cont        : num [1:16] 0.262 0.262 0.367 0.569 1.061 ...
 $ sri.cont            : num [1:16] -0.314 -1.308 -0.87 0.754 -0.746 ...
 $ tri.cont            : num [1:16] 1.725 1.526 1.289 -0.803 1.523 ...
 $ tcws.cont           : num [1:16] 1.525 1.716 1.716 -0.101 0.67 ...
 $ shrub_cov.cont      : num [1:16] 1.417 0.405 1.518 -0.607 -0.101 ...
 $ gramin_cov.cont     : num [1:16] -0.195 -0.389 -0.582 -0.582 -0.582 ...
 $ compet.cont         : num [1:16] -0.727 -0.959 -1.741 -0.674 -0.495 ...
 $ N_cont              : int 16
 $ tempjja.tot         : num [1:69] -0.333 -0.286 -0.222 -0.396 -0.426 ...
 $ tempcont.tot        : num [1:69] -2.45 -2.52 -2.49 -2.41 -2.39 ...
 $ precipjja.tot       : num [1:69] 0.711 0.869 1.058 0.629 0.59 ...
 $ N_plotgroups        : int 69
 $ xhat_compet         : num [1:100] -2.7 -2.68 -2.65 -2.63 -2.6 ...
 $ xhat_shrub_cover    : num [1:100] -1.21 -1.17 -1.12 -1.08 -1.03 ...
 $ xhat_graminoid_cover: num [1:100] -0.582 -0.504 -0.426 -0.348 -0.27 ...
 $ xhat_sri            : num [1:100] -3.11 -3.06 -3.02 -2.98 -2.94 ...
 $ xhat_tri            : num [1:100] -1.72 -1.66 -1.6 -1.54 -1.48 ...
 $ xhat_tcws           : num [1:100] -3.08 -3.03 -2.97 -2.92 -2.87 ...
 $ xhat_tempjja        : num [1:100] -2.57 -2.52 -2.48 -2.44 -2.4 ...
 $ xhat_precipjja      : num [1:100] -1.66 -1.62 -1.58 -1.55 -1.51 ...
 $ xhat_tempcont       : num [1:100] -2.52 -2.49 -2.45 -2.42 -2.38 ...
 $ Nxhat               : num 100
 $ xhat_tcws2          : num [1:2] -2.01 1.31
 $ Nxhat2              : num 2

# EmpNig ----
shrub_gradient_jags.EmpNig.data <- list(
  
  # plot level predictors, for discrete...
  cov.dis = EmpNig.dis$cover,
  plotgroup.dis = EmpNig.dis$plotgroup.NUM,
  tempjja.dis = EmpNig.dis$tempjjaC,
  sri.dis = EmpNig.dis$sriC,
  tri.dis = EmpNig.dis$triC,
  tcws.dis = EmpNig.dis$tcwsC,
  shrub_cov.dis = EmpNig.dis$EmpNig_shrub_coverC,
  gramin_cov.dis = EmpNig.dis$graminoid_coverC,
  compet.dis = EmpNig.dis$competC,
  N_discrete = nrow(EmpNig.dis),
  
  # ...and continuous part of the data
  cov.cont = EmpNig.cont$cover,
  plotgroup.cont = EmpNig.cont$plotgroup.NUM,
  tempjja.cont = EmpNig.cont$tempjjaC,
  sri.cont = EmpNig.cont$sriC,
  tri.cont = EmpNig.cont$triC,
  tcws.cont = EmpNig.cont$tcwsC,
  shrub_cov.cont = EmpNig.cont$EmpNig_shrub_coverC,
  gramin_cov.cont = EmpNig.cont$graminoid_coverC,
  compet.cont = EmpNig.cont$competC,
  N_cont = nrow(EmpNig.cont),
  
  # plot group level predictors
  tempjja.tot = EmpNig.tot %>% group_by(plotgroup.NUM) %>% summarise(tempjja.tot = mean(tempjjaC)) %>% pull(tempjja.tot), # one value per tXpg
  tempcont.tot = EmpNig.tot %>% group_by(plotgroup.NUM) %>% summarise(tempcont.tot = mean(tempcontC)) %>% pull(tempcont.tot),
  precipjja.tot = EmpNig.tot %>% group_by(plotgroup.NUM) %>% summarise(precipjja.tot = mean(precipjjaC)) %>% pull(precipjja.tot),
  N_plotgroups = length(unique(EmpNig.tot$site_alt_plotgroup_id)),
  
  # subset of values for prediction, for each predictor...
  xhat_compet = seq(from = min(EmpNig.tot$competC), to = max(EmpNig.tot$competC), length.out = 100),
  xhat_shrub_cover = seq(from = min(EmpNig.tot$EmpNig_shrub_coverC), to = max(EmpNig.tot$EmpNig_shrub_coverC), length.out = 100),
  xhat_graminoid_cover = seq(from = min(EmpNig.tot$graminoid_coverC), to = max(EmpNig.tot$graminoid_coverC), length.out = 100),
  xhat_sri = seq(from = min(EmpNig.tot$sriC), to = max(EmpNig.tot$sriC), length.out = 100),
  xhat_tri = seq(from = min(EmpNig.tot$triC), to = max(EmpNig.tot$triC), length.out = 100),
  xhat_tcws = seq(from = min(EmpNig.tot$tcwsC), to = max(EmpNig.tot$tcwsC), length.out = 100),
  xhat_tempjja = seq(from = min(EmpNig.tot$tempjjaC), to = max(EmpNig.tot$tempjjaC), length.out = 100),
  xhat_precipjja = seq(from = min(EmpNig.tot$precipjjaC), to = max(EmpNig.tot$precipjjaC), length.out = 100),
  xhat_tempcont = seq(from = min(EmpNig.tot$tempcontC), to = max(EmpNig.tot$tempcontC), length.out = 100),
  Nxhat = 100,
  
  # ... and for predicting at high/low temperature levels
  xhat_tcws2 = as.numeric(c(quantile(EmpNig.tot$tempjjaC,0.05),quantile(EmpNig.tot$tempjjaC,0.95))), 
  Nxhat2 = 2
)

`summarise()` ungrouping output (override with `.groups` argument)
`summarise()` ungrouping output (override with `.groups` argument)
`summarise()` ungrouping output (override with `.groups` argument)

str(shrub_gradient_jags.EmpNig.data)

List of 36
 $ cov.dis             : num [1:229] 0 0 0 0 0 0 0 0 0 0 ...
 $ plotgroup.dis       : num [1:229] 2 2 3 5 5 6 6 6 7 7 ...
 $ tempjja.dis         : num [1:229] -0.283 -0.288 -0.226 -0.42 -0.44 ...
 $ sri.dis             : num [1:229] 0.9453 -1.4348 -1.4755 0.7837 -0.0244 ...
 $ tri.dis             : num [1:229] -0.987 0.1646 0.0108 0.0788 0.2101 ...
 $ tcws.dis            : num [1:229] -1.935 -2.034 -0.366 -0.12 -0.466 ...
 $ shrub_cov.dis       : num [1:229] -0.985 -0.985 -0.985 -0.985 -0.985 ...
 $ gramin_cov.dis      : num [1:229] -0.582 -0.582 -0.582 -0.582 -0.582 ...
 $ compet.dis          : num [1:229] -0.37 -0.37 -0.37 -0.37 -0.37 ...
 $ N_discrete          : int 229
 $ cov.cont            : num [1:185] 0.52 0.84 0.6 0.64 0.52 0.16 0.16 0.64 0.16 0.52 ...
 $ plotgroup.cont      : num [1:185] 1 1 1 1 1 1 2 2 2 2 ...
 $ tempjja.cont        : num [1:185] -0.321 -0.321 -0.321 -0.345 -0.345 ...
 $ sri.cont            : num [1:185] 0.51 0.829 1.034 0.37 0.573 ...
 $ tri.cont            : num [1:185] -0.255 -0.314 -0.307 -0.274 -0.18 ...
 $ tcws.cont           : num [1:185] -0.4925 -0.0586 -0.0586 -0.0999 -0.6658 ...
 $ shrub_cov.cont      : num [1:185] -0.985 -0.985 -0.985 -0.745 -0.985 ...
 $ gramin_cov.cont     : num [1:185] -0.582 -0.389 -0.582 -0.582 -0.582 ...
 $ compet.cont         : num [1:185] -0.37 -0.37 -0.37 -0.515 -0.37 ...
 $ N_cont              : int 185
 $ tempjja.tot         : num [1:69] -0.333 -0.286 -0.222 -0.396 -0.426 ...
 $ tempcont.tot        : num [1:69] -2.45 -2.52 -2.49 -2.41 -2.39 ...
 $ precipjja.tot       : num [1:69] 0.711 0.869 1.058 0.629 0.59 ...
 $ N_plotgroups        : int 69
 $ xhat_compet         : num [1:100] -2.81 -2.78 -2.76 -2.73 -2.71 ...
 $ xhat_shrub_cover    : num [1:100] -0.985 -0.94 -0.895 -0.85 -0.806 ...
 $ xhat_graminoid_cover: num [1:100] -0.582 -0.504 -0.426 -0.348 -0.27 ...
 $ xhat_sri            : num [1:100] -3.11 -3.06 -3.02 -2.98 -2.94 ...
 $ xhat_tri            : num [1:100] -1.72 -1.66 -1.6 -1.54 -1.48 ...
 $ xhat_tcws           : num [1:100] -3.08 -3.03 -2.97 -2.92 -2.87 ...
 $ xhat_tempjja        : num [1:100] -2.57 -2.52 -2.48 -2.44 -2.4 ...
 $ xhat_precipjja      : num [1:100] -1.66 -1.62 -1.58 -1.55 -1.51 ...
 $ xhat_tempcont       : num [1:100] -2.52 -2.49 -2.45 -2.42 -2.38 ...
 $ Nxhat               : num 100
 $ xhat_tcws2          : num [1:2] -2.01 1.31
 $ Nxhat2              : num 2

# PhyCae ----
shrub_gradient_jags.PhyCae.data <- list(
  
  # plot level predictors, for discrete...
  cov.dis = PhyCae.dis$cover,
  plotgroup.dis = PhyCae.dis$plotgroup.NUM,
  tempjja.dis = PhyCae.dis$tempjjaC,
  sri.dis = PhyCae.dis$sriC,
  tri.dis = PhyCae.dis$triC,
  tcws.dis = PhyCae.dis$tcwsC,
  shrub_cov.dis = PhyCae.dis$PhyCae_shrub_coverC,
  gramin_cov.dis = PhyCae.dis$graminoid_coverC,
  compet.dis = PhyCae.dis$competC,
  N_discrete = nrow(PhyCae.dis),
  
  # ...and continuous part of the data
  cov.cont = PhyCae.cont$cover,
  plotgroup.cont = PhyCae.cont$plotgroup.NUM,
  tempjja.cont = PhyCae.cont$tempjjaC,
  sri.cont = PhyCae.cont$sriC,
  tri.cont = PhyCae.cont$triC,
  tcws.cont = PhyCae.cont$tcwsC,
  shrub_cov.cont = PhyCae.cont$PhyCae_shrub_coverC,
  gramin_cov.cont = PhyCae.cont$graminoid_coverC,
  compet.cont = PhyCae.cont$competC,
  N_cont = nrow(PhyCae.cont),
  
  # plot group level predictors
  tempjja.tot = PhyCae.tot %>% group_by(plotgroup.NUM) %>% summarise(tempjja.tot = mean(tempjjaC)) %>% pull(tempjja.tot), # one value per tXpg
  tempcont.tot = PhyCae.tot %>% group_by(plotgroup.NUM) %>% summarise(tempcont.tot = mean(tempcontC)) %>% pull(tempcont.tot),
  precipjja.tot = PhyCae.tot %>% group_by(plotgroup.NUM) %>% summarise(precipjja.tot = mean(precipjjaC)) %>% pull(precipjja.tot),
  N_plotgroups = length(unique(PhyCae.tot$site_alt_plotgroup_id)),
  
  # subset of values for prediction, for each predictor...
  xhat_compet = seq(from = min(PhyCae.tot$competC), to = max(PhyCae.tot$competC), length.out = 100),
  xhat_shrub_cover = seq(from = min(PhyCae.tot$PhyCae_shrub_coverC), to = max(PhyCae.tot$PhyCae_shrub_coverC), length.out = 100),
  xhat_graminoid_cover = seq(from = min(PhyCae.tot$graminoid_coverC), to = max(PhyCae.tot$graminoid_coverC), length.out = 100),
  xhat_sri = seq(from = min(PhyCae.tot$sriC), to = max(PhyCae.tot$sriC), length.out = 100),
  xhat_tri = seq(from = min(PhyCae.tot$triC), to = max(PhyCae.tot$triC), length.out = 100),
  xhat_tcws = seq(from = min(PhyCae.tot$tcwsC), to = max(PhyCae.tot$tcwsC), length.out = 100),
  xhat_tempjja = seq(from = min(PhyCae.tot$tempjjaC), to = max(PhyCae.tot$tempjjaC), length.out = 100),
  xhat_precipjja = seq(from = min(PhyCae.tot$precipjjaC), to = max(PhyCae.tot$precipjjaC), length.out = 100),
  xhat_tempcont = seq(from = min(PhyCae.tot$tempcontC), to = max(PhyCae.tot$tempcontC), length.out = 100),
  Nxhat = 100,
  
  # ... and for predicting at high/low temperature levels
  xhat_tcws2 = as.numeric(c(quantile(PhyCae.tot$tempjjaC,0.05),quantile(PhyCae.tot$tempjjaC,0.95))), 
  Nxhat2 = 2
)

`summarise()` ungrouping output (override with `.groups` argument)
`summarise()` ungrouping output (override with `.groups` argument)
`summarise()` ungrouping output (override with `.groups` argument)

str(shrub_gradient_jags.PhyCae.data)

List of 36
 $ cov.dis             : num [1:404] 0 0 0 0 0 0 0 0 0 0 ...
 $ plotgroup.dis       : num [1:404] 1 1 1 1 1 1 2 2 2 2 ...
 $ tempjja.dis         : num [1:404] -0.321 -0.321 -0.321 -0.345 -0.345 ...
 $ sri.dis             : num [1:404] 0.51 0.829 1.034 0.37 0.573 ...
 $ tri.dis             : num [1:404] -0.255 -0.314 -0.307 -0.274 -0.18 ...
 $ tcws.dis            : num [1:404] -0.4925 -0.0586 -0.0586 -0.0999 -0.6658 ...
 $ shrub_cov.dis       : num [1:404] 0.0921 0.8948 0.2928 0.5938 0.0921 ...
 $ gramin_cov.dis      : num [1:404] -0.582 -0.389 -0.582 -0.582 -0.582 ...
 $ compet.dis          : num [1:404] -0.118 -0.118 -0.118 -0.263 -0.118 ...
 $ N_discrete          : int 404
 $ cov.cont            : num [1:10] 0.04 0.08 0.08 0.12 0.08 0.04 0.04 0.12 0.44 0.12
 $ plotgroup.cont      : num [1:10] 9 11 14 17 24 30 36 51 51 69
 $ tempjja.cont        : num [1:10] 0.0508 0.7561 0.7023 1.0077 1.3121 ...
 $ sri.cont            : num [1:10] -0.7393 -0.0933 -0.2978 0.4817 0.075 ...
 $ tri.cont            : num [1:10] -1.116 -0.199 0.631 -0.24 0.115 ...
 $ tcws.cont           : num [1:10] 0.0921 1.162 -0.8479 0.7667 -0.3658 ...
 $ shrub_cov.cont      : num [1:10] -1.21224 2.19909 -0.40957 1.19576 -0.00824 ...
 $ gramin_cov.cont     : num [1:10] -0.38881 -0.00187 -0.19534 -0.19534 -0.38881 ...
 $ compet.cont         : num [1:10] -0.118 -1.419 -0.275 -0.768 -0.458 ...
 $ N_cont              : int 10
 $ tempjja.tot         : num [1:69] -0.333 -0.286 -0.222 -0.396 -0.426 ...
 $ tempcont.tot        : num [1:69] -2.45 -2.52 -2.49 -2.41 -2.39 ...
 $ precipjja.tot       : num [1:69] 0.711 0.869 1.058 0.629 0.59 ...
 $ N_plotgroups        : int 69
 $ xhat_compet         : num [1:100] -2.55 -2.53 -2.5 -2.48 -2.45 ...
 $ xhat_shrub_cover    : num [1:100] -1.21 -1.17 -1.12 -1.08 -1.03 ...
 $ xhat_graminoid_cover: num [1:100] -0.582 -0.504 -0.426 -0.348 -0.27 ...
 $ xhat_sri            : num [1:100] -3.11 -3.06 -3.02 -2.98 -2.94 ...
 $ xhat_tri            : num [1:100] -1.72 -1.66 -1.6 -1.54 -1.48 ...
 $ xhat_tcws           : num [1:100] -3.08 -3.03 -2.97 -2.92 -2.87 ...
 $ xhat_tempjja        : num [1:100] -2.57 -2.52 -2.48 -2.44 -2.4 ...
 $ xhat_precipjja      : num [1:100] -1.66 -1.62 -1.58 -1.55 -1.51 ...
 $ xhat_tempcont       : num [1:100] -2.52 -2.49 -2.45 -2.42 -2.38 ...
 $ Nxhat               : num 100
 $ xhat_tcws2          : num [1:2] -2.01 1.31
 $ Nxhat2              : num 2

# RhoGro ----
shrub_gradient_jags.RhoGro.data <- list(
  
  # plot level predictors, for discrete...
  cov.dis = RhoGro.dis$cover,
  plotgroup.dis = RhoGro.dis$plotgroup.NUM,
  tempjja.dis = RhoGro.dis$tempjjaC,
  sri.dis = RhoGro.dis$sriC,
  tri.dis = RhoGro.dis$triC,
  tcws.dis = RhoGro.dis$tcwsC,
  shrub_cov.dis = RhoGro.dis$RhoGro_shrub_coverC,
  gramin_cov.dis = RhoGro.dis$graminoid_coverC,
  compet.dis = RhoGro.dis$competC,
  N_discrete = nrow(RhoGro.dis),
  
  # ...and continuous part of the data
  cov.cont = RhoGro.cont$cover,
  plotgroup.cont = RhoGro.cont$plotgroup.NUM,
  tempjja.cont = RhoGro.cont$tempjjaC,
  sri.cont = RhoGro.cont$sriC,
  tri.cont = RhoGro.cont$triC,
  tcws.cont = RhoGro.cont$tcwsC,
  shrub_cov.cont = RhoGro.cont$RhoGro_shrub_coverC,
  gramin_cov.cont = RhoGro.cont$graminoid_coverC,
  compet.cont = RhoGro.cont$competC,
  N_cont = nrow(RhoGro.cont),
  
  # plot group level predictors
  tempjja.tot = RhoGro.tot %>% group_by(plotgroup.NUM) %>% summarise(tempjja.tot = mean(tempjjaC)) %>% pull(tempjja.tot), # one value per tXpg
  tempcont.tot = RhoGro.tot %>% group_by(plotgroup.NUM) %>% summarise(tempcont.tot = mean(tempcontC)) %>% pull(tempcont.tot),
  precipjja.tot = RhoGro.tot %>% group_by(plotgroup.NUM) %>% summarise(precipjja.tot = mean(precipjjaC)) %>% pull(precipjja.tot),
  N_plotgroups = length(unique(RhoGro.tot$site_alt_plotgroup_id)),
  
  # subset of values for prediction, for each predictor...
  xhat_compet = seq(from = min(RhoGro.tot$competC), to = max(RhoGro.tot$competC), length.out = 100),
  xhat_shrub_cover = seq(from = min(RhoGro.tot$RhoGro_shrub_coverC), to = max(RhoGro.tot$RhoGro_shrub_coverC), length.out = 100),
  xhat_graminoid_cover = seq(from = min(RhoGro.tot$graminoid_coverC), to = max(RhoGro.tot$graminoid_coverC), length.out = 100),
  xhat_sri = seq(from = min(RhoGro.tot$sriC), to = max(RhoGro.tot$sriC), length.out = 100),
  xhat_tri = seq(from = min(RhoGro.tot$triC), to = max(RhoGro.tot$triC), length.out = 100),
  xhat_tcws = seq(from = min(RhoGro.tot$tcwsC), to = max(RhoGro.tot$tcwsC), length.out = 100),
  xhat_tempjja = seq(from = min(RhoGro.tot$tempjjaC), to = max(RhoGro.tot$tempjjaC), length.out = 100),
  xhat_precipjja = seq(from = min(RhoGro.tot$precipjjaC), to = max(RhoGro.tot$precipjjaC), length.out = 100),
  xhat_tempcont = seq(from = min(RhoGro.tot$tempcontC), to = max(RhoGro.tot$tempcontC), length.out = 100),
  Nxhat = 100,
  
  # ... and for predicting at high/low temperature levels
  xhat_tcws2 = as.numeric(c(quantile(RhoGro.tot$tempjjaC,0.05),quantile(RhoGro.tot$tempjjaC,0.95))), 
  Nxhat2 = 2
)

`summarise()` ungrouping output (override with `.groups` argument)
`summarise()` ungrouping output (override with `.groups` argument)
`summarise()` ungrouping output (override with `.groups` argument)

str(shrub_gradient_jags.RhoGro.data)

List of 36
 $ cov.dis             : num [1:325] 0 0 0 0 0 0 0 0 0 0 ...
 $ plotgroup.dis       : num [1:325] 1 1 1 1 1 1 2 2 2 2 ...
 $ tempjja.dis         : num [1:325] -0.321 -0.321 -0.321 -0.345 -0.345 ...
 $ sri.dis             : num [1:325] 0.51 0.829 1.034 0.37 0.573 ...
 $ tri.dis             : num [1:325] -0.255 -0.314 -0.307 -0.274 -0.18 ...
 $ tcws.dis            : num [1:325] -0.4925 -0.0586 -0.0586 -0.0999 -0.6658 ...
 $ shrub_cov.dis       : num [1:325] 0.24 1.138 0.464 0.801 0.24 ...
 $ gramin_cov.dis      : num [1:325] -0.582 -0.389 -0.582 -0.582 -0.582 ...
 $ compet.dis          : num [1:325] -0.0764 -0.0764 -0.0764 -0.221 -0.0764 ...
 $ N_discrete          : int 325
 $ cov.cont            : num [1:89] 0.08 0.12 0.52 0.36 0.28 0.04 0.04 0.16 0.04 0.04 ...
 $ plotgroup.cont      : num [1:89] 5 11 20 20 20 20 20 20 24 24 ...
 $ tempjja.cont        : num [1:89] -0.42 0.756 0.374 0.426 0.426 ...
 $ sri.cont            : num [1:89] 0.8231 -0.9499 -0.0661 -0.0445 -2.5789 ...
 $ tri.cont            : num [1:89] 0.4531 -0.0842 1.311 1.4128 0.9228 ...
 $ tcws.cont           : num [1:89] 0.176 1.052 0.586 0.699 0.839 ...
 $ shrub_cov.cont      : num [1:89] 2.261 1.026 0.689 0.24 0.689 ...
 $ gramin_cov.cont     : num [1:89] 0.965 -0.582 -0.582 -0.582 -0.195 ...
 $ compet.cont         : num [1:89] -1.033 -0.74 -1.233 -0.961 -1.353 ...
 $ N_cont              : int 89
 $ tempjja.tot         : num [1:69] -0.333 -0.286 -0.222 -0.396 -0.426 ...
 $ tempcont.tot        : num [1:69] -2.45 -2.52 -2.49 -2.41 -2.39 ...
 $ precipjja.tot       : num [1:69] 0.711 0.869 1.058 0.629 0.59 ...
 $ N_plotgroups        : int 69
 $ xhat_compet         : num [1:100] -2.48 -2.45 -2.43 -2.4 -2.38 ...
 $ xhat_shrub_cover    : num [1:100] -1.22 -1.17 -1.13 -1.08 -1.04 ...
 $ xhat_graminoid_cover: num [1:100] -0.582 -0.504 -0.426 -0.348 -0.27 ...
 $ xhat_sri            : num [1:100] -3.11 -3.06 -3.02 -2.98 -2.94 ...
 $ xhat_tri            : num [1:100] -1.72 -1.66 -1.6 -1.54 -1.48 ...
 $ xhat_tcws           : num [1:100] -3.08 -3.03 -2.97 -2.92 -2.87 ...
 $ xhat_tempjja        : num [1:100] -2.57 -2.52 -2.48 -2.44 -2.4 ...
 $ xhat_precipjja      : num [1:100] -1.66 -1.62 -1.58 -1.55 -1.51 ...
 $ xhat_tempcont       : num [1:100] -2.52 -2.49 -2.45 -2.42 -2.38 ...
 $ Nxhat               : num 100
 $ xhat_tcws2          : num [1:2] -2.01 1.31
 $ Nxhat2              : num 2

# RhoTom ----
shrub_gradient_jags.RhoTom.data <- list(
  
  # plot level predictors, for discrete...
  cov.dis = RhoTom.dis$cover,
  plotgroup.dis = RhoTom.dis$plotgroup.NUM,
  tempjja.dis = RhoTom.dis$tempjjaC,
  sri.dis = RhoTom.dis$sriC,
  tri.dis = RhoTom.dis$triC,
  tcws.dis = RhoTom.dis$tcwsC,
  shrub_cov.dis = RhoTom.dis$RhoTom_shrub_coverC,
  gramin_cov.dis = RhoTom.dis$graminoid_coverC,
  compet.dis = RhoTom.dis$competC,
  N_discrete = nrow(RhoTom.dis),
  
  # ...and continuous part of the data
  cov.cont = RhoTom.cont$cover,
  plotgroup.cont = RhoTom.cont$plotgroup.NUM,
  tempjja.cont = RhoTom.cont$tempjjaC,
  sri.cont = RhoTom.cont$sriC,
  tri.cont = RhoTom.cont$triC,
  tcws.cont = RhoTom.cont$tcwsC,
  shrub_cov.cont = RhoTom.cont$RhoTom_shrub_coverC,
  gramin_cov.cont = RhoTom.cont$graminoid_coverC,
  compet.cont = RhoTom.cont$competC,
  N_cont = nrow(RhoTom.cont),
  
  # plot group level predictors
  tempjja.tot = RhoTom.tot %>% group_by(plotgroup.NUM) %>% summarise(tempjja.tot = mean(tempjjaC)) %>% pull(tempjja.tot), # one value per tXpg
  tempcont.tot = RhoTom.tot %>% group_by(plotgroup.NUM) %>% summarise(tempcont.tot = mean(tempcontC)) %>% pull(tempcont.tot),
  precipjja.tot = RhoTom.tot %>% group_by(plotgroup.NUM) %>% summarise(precipjja.tot = mean(precipjjaC)) %>% pull(precipjja.tot),
  N_plotgroups = length(unique(RhoTom.tot$site_alt_plotgroup_id)),
  
  # subset of values for prediction, for each predictor...
  xhat_compet = seq(from = min(RhoTom.tot$competC), to = max(RhoTom.tot$competC), length.out = 100),
  xhat_shrub_cover = seq(from = min(RhoTom.tot$RhoTom_shrub_coverC), to = max(RhoTom.tot$RhoTom_shrub_coverC), length.out = 100),
  xhat_graminoid_cover = seq(from = min(RhoTom.tot$graminoid_coverC), to = max(RhoTom.tot$graminoid_coverC), length.out = 100),
  xhat_sri = seq(from = min(RhoTom.tot$sriC), to = max(RhoTom.tot$sriC), length.out = 100),
  xhat_tri = seq(from = min(RhoTom.tot$triC), to = max(RhoTom.tot$triC), length.out = 100),
  xhat_tcws = seq(from = min(RhoTom.tot$tcwsC), to = max(RhoTom.tot$tcwsC), length.out = 100),
  xhat_tempjja = seq(from = min(RhoTom.tot$tempjjaC), to = max(RhoTom.tot$tempjjaC), length.out = 100),
  xhat_precipjja = seq(from = min(RhoTom.tot$precipjjaC), to = max(RhoTom.tot$precipjjaC), length.out = 100),
  xhat_tempcont = seq(from = min(RhoTom.tot$tempcontC), to = max(RhoTom.tot$tempcontC), length.out = 100),
  Nxhat = 100,
  
  # ... and for predicting at high/low temperature levels
  xhat_tcws2 = as.numeric(c(quantile(RhoTom.tot$tempjjaC,0.05),quantile(RhoTom.tot$tempjjaC,0.95))), 
  Nxhat2 = 2
)

`summarise()` ungrouping output (override with `.groups` argument)
`summarise()` ungrouping output (override with `.groups` argument)
`summarise()` ungrouping output (override with `.groups` argument)

str(shrub_gradient_jags.RhoTom.data)

List of 36
 $ cov.dis             : num [1:401] 0 0 0 0 0 0 0 0 0 0 ...
 $ plotgroup.dis       : num [1:401] 1 1 1 1 1 1 2 2 2 2 ...
 $ tempjja.dis         : num [1:401] -0.321 -0.321 -0.321 -0.345 -0.345 ...
 $ sri.dis             : num [1:401] 0.51 0.829 1.034 0.37 0.573 ...
 $ tri.dis             : num [1:401] -0.255 -0.314 -0.307 -0.274 -0.18 ...
 $ tcws.dis            : num [1:401] -0.4925 -0.0586 -0.0586 -0.0999 -0.6658 ...
 $ shrub_cov.dis       : num [1:401] 0.0983 0.9077 0.3006 0.6042 0.0983 ...
 $ gramin_cov.dis      : num [1:401] -0.582 -0.389 -0.582 -0.582 -0.582 ...
 $ compet.dis          : num [1:401] -0.0764 -0.0764 -0.0764 -0.221 -0.0764 ...
 $ N_discrete          : int 401
 $ cov.cont            : num [1:13] 0.12 0.2 0.12 0.4 0.2 0.2 0.2 0.24 0.04 0.2 ...
 $ plotgroup.cont      : num [1:13] 11 11 12 30 30 36 39 39 42 43 ...
 $ tempjja.cont        : num [1:13] 0.756 0.756 0.542 0.148 0.262 ...
 $ sri.cont            : num [1:13] -0.95 0.723 -0.185 0.633 -1.308 ...
 $ tri.cont            : num [1:13] -0.0842 -0.2235 -0.422 1.6901 1.5259 ...
 $ tcws.cont           : num [1:13] 1.05 1.05 0.74 1.72 1.72 ...
 $ shrub_cov.cont      : num [1:13] 0.80655 0.70536 1.21129 0.30062 -0.00293 ...
 $ gramin_cov.cont     : num [1:13] -0.582 -0.582 1.546 -0.582 -0.389 ...
 $ compet.cont         : num [1:13] -0.74 -0.51 -0.378 -0.565 -0.716 ...
 $ N_cont              : int 13
 $ tempjja.tot         : num [1:69] -0.333 -0.286 -0.222 -0.396 -0.426 ...
 $ tempcont.tot        : num [1:69] -2.45 -2.52 -2.49 -2.41 -2.39 ...
 $ precipjja.tot       : num [1:69] 0.711 0.869 1.058 0.629 0.59 ...
 $ N_plotgroups        : int 69
 $ xhat_compet         : num [1:100] -2.48 -2.45 -2.43 -2.4 -2.38 ...
 $ xhat_shrub_cover    : num [1:100] -1.22 -1.17 -1.13 -1.08 -1.03 ...
 $ xhat_graminoid_cover: num [1:100] -0.582 -0.504 -0.426 -0.348 -0.27 ...
 $ xhat_sri            : num [1:100] -3.11 -3.06 -3.02 -2.98 -2.94 ...
 $ xhat_tri            : num [1:100] -1.72 -1.66 -1.6 -1.54 -1.48 ...
 $ xhat_tcws           : num [1:100] -3.08 -3.03 -2.97 -2.92 -2.87 ...
 $ xhat_tempjja        : num [1:100] -2.57 -2.52 -2.48 -2.44 -2.4 ...
 $ xhat_precipjja      : num [1:100] -1.66 -1.62 -1.58 -1.55 -1.51 ...
 $ xhat_tempcont       : num [1:100] -2.52 -2.49 -2.45 -2.42 -2.38 ...
 $ Nxhat               : num 100
 $ xhat_tcws2          : num [1:2] -2.01 1.31
 $ Nxhat2              : num 2

# SalArc ----
shrub_gradient_jags.SalArc.data <- list(
  
  # plot level predictors, for discrete...
  cov.dis = SalArc.dis$cover,
  plotgroup.dis = SalArc.dis$plotgroup.NUM,
  tempjja.dis = SalArc.dis$tempjjaC,
  sri.dis = SalArc.dis$sriC,
  tri.dis = SalArc.dis$triC,
  tcws.dis = SalArc.dis$tcwsC,
  shrub_cov.dis = SalArc.dis$SalArc_shrub_coverC,
  gramin_cov.dis = SalArc.dis$graminoid_coverC,
  compet.dis = SalArc.dis$competC,
  N_discrete = nrow(SalArc.dis),
  
  # ...and continuous part of the data
  cov.cont = SalArc.cont$cover,
  plotgroup.cont = SalArc.cont$plotgroup.NUM,
  tempjja.cont = SalArc.cont$tempjjaC,
  sri.cont = SalArc.cont$sriC,
  tri.cont = SalArc.cont$triC,
  tcws.cont = SalArc.cont$tcwsC,
  shrub_cov.cont = SalArc.cont$SalArc_shrub_coverC,
  gramin_cov.cont = SalArc.cont$graminoid_coverC,
  compet.cont = SalArc.cont$competC,
  N_cont = nrow(SalArc.cont),
  
  # plot group level predictors
  tempjja.tot = SalArc.tot %>% group_by(plotgroup.NUM) %>% summarise(tempjja.tot = mean(tempjjaC)) %>% pull(tempjja.tot), # one value per tXpg
  tempcont.tot = SalArc.tot %>% group_by(plotgroup.NUM) %>% summarise(tempcont.tot = mean(tempcontC)) %>% pull(tempcont.tot),
  precipjja.tot = SalArc.tot %>% group_by(plotgroup.NUM) %>% summarise(precipjja.tot = mean(precipjjaC)) %>% pull(precipjja.tot),
  N_plotgroups = length(unique(SalArc.tot$site_alt_plotgroup_id)),
  
  # subset of values for prediction, for each predictor...
  xhat_compet = seq(from = min(SalArc.tot$competC), to = max(SalArc.tot$competC), length.out = 100),
  xhat_shrub_cover = seq(from = min(SalArc.tot$SalArc_shrub_coverC), to = max(SalArc.tot$SalArc_shrub_coverC), length.out = 100),
  xhat_graminoid_cover = seq(from = min(SalArc.tot$graminoid_coverC), to = max(SalArc.tot$graminoid_coverC), length.out = 100),
  xhat_sri = seq(from = min(SalArc.tot$sriC), to = max(SalArc.tot$sriC), length.out = 100),
  xhat_tri = seq(from = min(SalArc.tot$triC), to = max(SalArc.tot$triC), length.out = 100),
  xhat_tcws = seq(from = min(SalArc.tot$tcwsC), to = max(SalArc.tot$tcwsC), length.out = 100),
  xhat_tempjja = seq(from = min(SalArc.tot$tempjjaC), to = max(SalArc.tot$tempjjaC), length.out = 100),
  xhat_precipjja = seq(from = min(SalArc.tot$precipjjaC), to = max(SalArc.tot$precipjjaC), length.out = 100),
  xhat_tempcont = seq(from = min(SalArc.tot$tempcontC), to = max(SalArc.tot$tempcontC), length.out = 100),
  Nxhat = 100,
  
  # ... and for predicting at high/low temperature levels
  xhat_tcws2 = as.numeric(c(quantile(SalArc.tot$tempjjaC,0.05),quantile(SalArc.tot$tempjjaC,0.95))), 
  Nxhat2 = 2
)

`summarise()` ungrouping output (override with `.groups` argument)
`summarise()` ungrouping output (override with `.groups` argument)
`summarise()` ungrouping output (override with `.groups` argument)

str(shrub_gradient_jags.SalArc.data)

List of 36
 $ cov.dis             : num [1:401] 0 0 0 0 0 0 0 0 0 0 ...
 $ plotgroup.dis       : num [1:401] 1 1 1 1 1 1 2 2 2 2 ...
 $ tempjja.dis         : num [1:401] -0.321 -0.321 -0.321 -0.345 -0.345 ...
 $ sri.dis             : num [1:401] 0.51 0.829 1.034 0.37 0.573 ...
 $ tri.dis             : num [1:401] -0.255 -0.314 -0.307 -0.274 -0.18 ...
 $ tcws.dis            : num [1:401] -0.4925 -0.0586 -0.0586 -0.0999 -0.6658 ...
 $ shrub_cov.dis       : num [1:401] 0.09 0.894 0.291 0.592 0.09 ...
 $ gramin_cov.dis      : num [1:401] -0.582 -0.389 -0.582 -0.582 -0.582 ...
 $ compet.dis          : num [1:401] 1.76 1.76 1.76 1.61 1.76 ...
 $ N_discrete          : int 401
 $ cov.cont            : num [1:13] 0.04 0.04 0.08 0.04 0.04 0.04 0.04 0.2 0.08 0.04 ...
 $ plotgroup.cont      : num [1:13] 4 4 5 9 16 17 18 30 43 45 ...
 $ tempjja.cont        : num [1:13] -0.3953 -0.3953 -0.4199 0.0508 1.3027 ...
 $ sri.cont            : num [1:13] -1.344 -2.825 0.823 0.14 0.305 ...
 $ tri.cont            : num [1:13] 1.583 1.364 0.453 -0.511 -0.944 ...
 $ tcws.cont           : num [1:13] 0.248 0.163 0.176 0.182 -2.3 ...
 $ shrub_cov.cont      : num [1:13] 0.291 0.09 1.898 -0.111 -0.714 ...
 $ gramin_cov.cont     : num [1:13] -0.582 -0.582 0.965 0.772 -0.582 ...
 $ compet.cont         : num [1:13] 1.758 1.18 0.948 1.758 1.397 ...
 $ N_cont              : int 13
 $ tempjja.tot         : num [1:69] -0.333 -0.286 -0.222 -0.396 -0.426 ...
 $ tempcont.tot        : num [1:69] -2.45 -2.52 -2.49 -2.41 -2.39 ...
 $ precipjja.tot       : num [1:69] 0.711 0.869 1.058 0.629 0.59 ...
 $ N_plotgroups        : int 69
 $ xhat_compet         : num [1:100] -0.37 -0.349 -0.327 -0.306 -0.284 ...
 $ xhat_shrub_cover    : num [1:100] -1.22 -1.17 -1.12 -1.08 -1.03 ...
 $ xhat_graminoid_cover: num [1:100] -0.582 -0.504 -0.426 -0.348 -0.27 ...
 $ xhat_sri            : num [1:100] -3.11 -3.06 -3.02 -2.98 -2.94 ...
 $ xhat_tri            : num [1:100] -1.72 -1.66 -1.6 -1.54 -1.48 ...
 $ xhat_tcws           : num [1:100] -3.08 -3.03 -2.97 -2.92 -2.87 ...
 $ xhat_tempjja        : num [1:100] -2.57 -2.52 -2.48 -2.44 -2.4 ...
 $ xhat_precipjja      : num [1:100] -1.66 -1.62 -1.58 -1.55 -1.51 ...
 $ xhat_tempcont       : num [1:100] -2.52 -2.49 -2.45 -2.42 -2.38 ...
 $ Nxhat               : num 100
 $ xhat_tcws2          : num [1:2] -2.01 1.31
 $ Nxhat2              : num 2

# SalGla ----
shrub_gradient_jags.SalGla.data <- list(
  
  # plot level predictors, for discrete...
  cov.dis = SalGla.dis$cover,
  plotgroup.dis = SalGla.dis$plotgroup.NUM,
  tempjja.dis = SalGla.dis$tempjjaC,
  sri.dis = SalGla.dis$sriC,
  tri.dis = SalGla.dis$triC,
  tcws.dis = SalGla.dis$tcwsC,
  shrub_cov.dis = SalGla.dis$SalGla_shrub_coverC,
  gramin_cov.dis = SalGla.dis$graminoid_coverC,
  compet.dis = SalGla.dis$competC,
  N_discrete = nrow(SalGla.dis),
  
  # ...and continuous part of the data
  cov.cont = SalGla.cont$cover,
  plotgroup.cont = SalGla.cont$plotgroup.NUM,
  tempjja.cont = SalGla.cont$tempjjaC,
  sri.cont = SalGla.cont$sriC,
  tri.cont = SalGla.cont$triC,
  tcws.cont = SalGla.cont$tcwsC,
  shrub_cov.cont = SalGla.cont$SalGla_shrub_coverC,
  gramin_cov.cont = SalGla.cont$graminoid_coverC,
  compet.cont = SalGla.cont$competC,
  N_cont = nrow(SalGla.cont),
  
  # plot group level predictors
  tempjja.tot = SalGla.tot %>% group_by(plotgroup.NUM) %>% summarise(tempjja.tot = mean(tempjjaC)) %>% pull(tempjja.tot), # one value per tXpg
  tempcont.tot = SalGla.tot %>% group_by(plotgroup.NUM) %>% summarise(tempcont.tot = mean(tempcontC)) %>% pull(tempcont.tot),
  precipjja.tot = SalGla.tot %>% group_by(plotgroup.NUM) %>% summarise(precipjja.tot = mean(precipjjaC)) %>% pull(precipjja.tot),
  N_plotgroups = length(unique(SalGla.tot$site_alt_plotgroup_id)),
  
  # subset of values for prediction, for each predictor...
  xhat_compet = seq(from = min(SalGla.tot$competC), to = max(SalGla.tot$competC), length.out = 100),
  xhat_shrub_cover = seq(from = min(SalGla.tot$SalGla_shrub_coverC), to = max(SalGla.tot$SalGla_shrub_coverC), length.out = 100),
  xhat_graminoid_cover = seq(from = min(SalGla.tot$graminoid_coverC), to = max(SalGla.tot$graminoid_coverC), length.out = 100),
  xhat_sri = seq(from = min(SalGla.tot$sriC), to = max(SalGla.tot$sriC), length.out = 100),
  xhat_tri = seq(from = min(SalGla.tot$triC), to = max(SalGla.tot$triC), length.out = 100),
  xhat_tcws = seq(from = min(SalGla.tot$tcwsC), to = max(SalGla.tot$tcwsC), length.out = 100),
  xhat_tempjja = seq(from = min(SalGla.tot$tempjjaC), to = max(SalGla.tot$tempjjaC), length.out = 100),
  xhat_precipjja = seq(from = min(SalGla.tot$precipjjaC), to = max(SalGla.tot$precipjjaC), length.out = 100),
  xhat_tempcont = seq(from = min(SalGla.tot$tempcontC), to = max(SalGla.tot$tempcontC), length.out = 100),
  Nxhat = 100,
  
  # ... and for predicting at high/low temperature levels
  xhat_tcws2 = as.numeric(c(quantile(SalGla.tot$tempjjaC,0.05),quantile(SalGla.tot$tempjjaC,0.95))), 
  Nxhat2 = 2
)

`summarise()` ungrouping output (override with `.groups` argument)
`summarise()` ungrouping output (override with `.groups` argument)
`summarise()` ungrouping output (override with `.groups` argument)

str(shrub_gradient_jags.SalGla.data)

List of 36
 $ cov.dis             : num [1:289] 0 0 0 0 0 0 0 0 0 0 ...
 $ plotgroup.dis       : num [1:289] 1 1 1 1 1 1 2 2 2 2 ...
 $ tempjja.dis         : num [1:289] -0.321 -0.321 -0.321 -0.345 -0.345 ...
 $ sri.dis             : num [1:289] 0.51 0.829 1.034 0.37 0.573 ...
 $ tri.dis             : num [1:289] -0.255 -0.314 -0.307 -0.274 -0.18 ...
 $ tcws.dis            : num [1:289] -0.4925 -0.0586 -0.0586 -0.0999 -0.6658 ...
 $ shrub_cov.dis       : num [1:289] 0.291 1.163 0.509 0.836 0.291 ...
 $ gramin_cov.dis      : num [1:289] -0.582 -0.389 -0.582 -0.582 -0.582 ...
 $ compet.dis          : num [1:289] 1.76 1.76 1.76 1.61 1.76 ...
 $ N_discrete          : int 289
 $ cov.cont            : num [1:125] 0.32 0.12 0.2 0.08 0.44 0.08 0.04 0.12 0.04 0.04 ...
 $ plotgroup.cont      : num [1:125] 2 2 4 4 6 8 11 11 12 14 ...
 $ tempjja.cont        : num [1:125] -0.288 -0.288 -0.395 -0.395 -0.4 ...
 $ sri.cont            : num [1:125] -1.562 -0.758 0.646 -2.68 -0.409 ...
 $ tri.cont            : num [1:125] 0.000194 0.116122 -0.957191 1.458727 0.141645 ...
 $ tcws.cont           : num [1:125] -1.542 -1.229 0.431 0.163 -0.848 ...
 $ shrub_cov.cont      : num [1:125] -0.69 0.291 0.836 0.727 0.4 ...
 $ gramin_cov.cont     : num [1:125] -0.582 -0.582 -0.195 -0.582 -0.582 ...
 $ compet.cont         : num [1:125] 1.76 1.76 1.76 1.76 1.76 ...
 $ N_cont              : int 125
 $ tempjja.tot         : num [1:69] -0.333 -0.286 -0.222 -0.396 -0.426 ...
 $ tempcont.tot        : num [1:69] -2.45 -2.52 -2.49 -2.41 -2.39 ...
 $ precipjja.tot       : num [1:69] 0.711 0.869 1.058 0.629 0.59 ...
 $ N_plotgroups        : int 69
 $ xhat_compet         : num [1:100] -0.37 -0.349 -0.327 -0.306 -0.284 ...
 $ xhat_shrub_cover    : num [1:100] -1.126 -1.083 -1.04 -0.997 -0.954 ...
 $ xhat_graminoid_cover: num [1:100] -0.582 -0.504 -0.426 -0.348 -0.27 ...
 $ xhat_sri            : num [1:100] -3.11 -3.06 -3.02 -2.98 -2.94 ...
 $ xhat_tri            : num [1:100] -1.72 -1.66 -1.6 -1.54 -1.48 ...
 $ xhat_tcws           : num [1:100] -3.08 -3.03 -2.97 -2.92 -2.87 ...
 $ xhat_tempjja        : num [1:100] -2.57 -2.52 -2.48 -2.44 -2.4 ...
 $ xhat_precipjja      : num [1:100] -1.66 -1.62 -1.58 -1.55 -1.51 ...
 $ xhat_tempcont       : num [1:100] -2.52 -2.49 -2.45 -2.42 -2.38 ...
 $ Nxhat               : num 100
 $ xhat_tcws2          : num [1:2] -2.01 1.31
 $ Nxhat2              : num 2

# VacUli ----
shrub_gradient_jags.VacUli.data <- list(
  
  # plot level predictors, for discrete...
  cov.dis = VacUli.dis$cover,
  plotgroup.dis = VacUli.dis$plotgroup.NUM,
  tempjja.dis = VacUli.dis$tempjjaC,
  sri.dis = VacUli.dis$sriC,
  tri.dis = VacUli.dis$triC,
  tcws.dis = VacUli.dis$tcwsC,
  shrub_cov.dis = VacUli.dis$VacUli_shrub_coverC,
  gramin_cov.dis = VacUli.dis$graminoid_coverC,
  compet.dis = VacUli.dis$competC,
  N_discrete = nrow(VacUli.dis),
  
  # ...and continuous part of the data
  cov.cont = VacUli.cont$cover,
  plotgroup.cont = VacUli.cont$plotgroup.NUM,
  tempjja.cont = VacUli.cont$tempjjaC,
  sri.cont = VacUli.cont$sriC,
  tri.cont = VacUli.cont$triC,
  tcws.cont = VacUli.cont$tcwsC,
  shrub_cov.cont = VacUli.cont$VacUli_shrub_coverC,
  gramin_cov.cont = VacUli.cont$graminoid_coverC,
  compet.cont = VacUli.cont$competC,
  N_cont = nrow(VacUli.cont),
  
  # plot group level predictors
  tempjja.tot = VacUli.tot %>% group_by(plotgroup.NUM) %>% summarise(tempjja.tot = mean(tempjjaC)) %>% pull(tempjja.tot), # one value per tXpg
  tempcont.tot = VacUli.tot %>% group_by(plotgroup.NUM) %>% summarise(tempcont.tot = mean(tempcontC)) %>% pull(tempcont.tot),
  precipjja.tot = VacUli.tot %>% group_by(plotgroup.NUM) %>% summarise(precipjja.tot = mean(precipjjaC)) %>% pull(precipjja.tot),
  N_plotgroups = length(unique(VacUli.tot$site_alt_plotgroup_id)),
  
  # subset of values for prediction, for each predictor...
  xhat_compet = seq(from = min(VacUli.tot$competC), to = max(VacUli.tot$competC), length.out = 100),
  xhat_shrub_cover = seq(from = min(VacUli.tot$VacUli_shrub_coverC), to = max(VacUli.tot$VacUli_shrub_coverC), length.out = 100),
  xhat_graminoid_cover = seq(from = min(VacUli.tot$graminoid_coverC), to = max(VacUli.tot$graminoid_coverC), length.out = 100),
  xhat_sri = seq(from = min(VacUli.tot$sriC), to = max(VacUli.tot$sriC), length.out = 100),
  xhat_tri = seq(from = min(VacUli.tot$triC), to = max(VacUli.tot$triC), length.out = 100),
  xhat_tcws = seq(from = min(VacUli.tot$tcwsC), to = max(VacUli.tot$tcwsC), length.out = 100),
  xhat_tempjja = seq(from = min(VacUli.tot$tempjjaC), to = max(VacUli.tot$tempjjaC), length.out = 100),
  xhat_precipjja = seq(from = min(VacUli.tot$precipjjaC), to = max(VacUli.tot$precipjjaC), length.out = 100),
  xhat_tempcont = seq(from = min(VacUli.tot$tempcontC), to = max(VacUli.tot$tempcontC), length.out = 100),
  Nxhat = 100,
  
  # ... and for predicting at high/low temperature levels
  xhat_tcws2 = as.numeric(c(quantile(VacUli.tot$tempjjaC,0.05),quantile(VacUli.tot$tempjjaC,0.95))), 
  Nxhat2 = 2
)

`summarise()` ungrouping output (override with `.groups` argument)
`summarise()` ungrouping output (override with `.groups` argument)
`summarise()` ungrouping output (override with `.groups` argument)

str(shrub_gradient_jags.VacUli.data)

List of 36
 $ cov.dis             : num [1:309] 0 0 0 0 0 0 0 0 0 0 ...
 $ plotgroup.dis       : num [1:309] 1 1 1 1 2 2 2 2 2 2 ...
 $ tempjja.dis         : num [1:309] -0.321 -0.321 -0.321 -0.345 -0.283 ...
 $ sri.dis             : num [1:309] 0.51 0.829 1.034 0.573 0.945 ...
 $ tri.dis             : num [1:309] -0.255 -0.314 -0.307 -0.18 -0.987 ...
 $ tcws.dis            : num [1:309] -0.4925 -0.0586 -0.0586 -0.6658 -1.935 ...
 $ shrub_cov.dis       : num [1:309] 0.243 1.121 0.462 0.243 -1.185 ...
 $ gramin_cov.dis      : num [1:309] -0.582 -0.389 -0.582 -0.582 -0.582 ...
 $ compet.dis          : num [1:309] 1.44 1.44 1.44 1.44 -0.37 ...
 $ N_discrete          : int 309
 $ cov.cont            : num [1:105] 0.08 0.04 0.24 0.28 0.08 0.32 0.36 0.28 0.08 0.16 ...
 $ plotgroup.cont      : num [1:105] 1 1 3 3 4 4 5 10 10 11 ...
 $ tempjja.cont        : num [1:105] -0.345 -0.345 -0.226 -0.226 -0.395 ...
 $ sri.cont            : num [1:105] 0.37 0.688 -0.207 -1.955 -0.556 ...
 $ tri.cont            : num [1:105] -0.274 0.19 0.149 0.385 1.362 ...
 $ tcws.cont           : num [1:105] -0.0999 -0.6658 -0.2434 -0.2434 0.1631 ...
 $ shrub_cov.cont      : num [1:105] 0.572 -0.7454 0.1329 0.6818 0.0231 ...
 $ gramin_cov.cont     : num [1:105] -0.582 -0.582 -0.582 -0.582 -0.582 ...
 $ compet.cont         : num [1:105] 1.44 1.44 1.44 1.44 1.44 ...
 $ N_cont              : int 105
 $ tempjja.tot         : num [1:69] -0.333 -0.286 -0.222 -0.396 -0.426 ...
 $ tempcont.tot        : num [1:69] -2.45 -2.52 -2.49 -2.41 -2.39 ...
 $ precipjja.tot       : num [1:69] 0.711 0.869 1.058 0.629 0.59 ...
 $ N_plotgroups        : int 69
 $ xhat_compet         : num [1:100] -0.999 -0.974 -0.95 -0.925 -0.9 ...
 $ xhat_shrub_cover    : num [1:100] -1.185 -1.135 -1.085 -1.035 -0.985 ...
 $ xhat_graminoid_cover: num [1:100] -0.582 -0.504 -0.426 -0.348 -0.27 ...
 $ xhat_sri            : num [1:100] -3.11 -3.06 -3.02 -2.98 -2.94 ...
 $ xhat_tri            : num [1:100] -1.72 -1.66 -1.6 -1.54 -1.48 ...
 $ xhat_tcws           : num [1:100] -3.08 -3.03 -2.97 -2.92 -2.87 ...
 $ xhat_tempjja        : num [1:100] -2.57 -2.52 -2.48 -2.44 -2.4 ...
 $ xhat_precipjja      : num [1:100] -1.66 -1.62 -1.58 -1.55 -1.51 ...
 $ xhat_tempcont       : num [1:100] -2.52 -2.49 -2.45 -2.42 -2.38 ...
 $ Nxhat               : num 100
 $ xhat_tcws2          : num [1:2] -2.01 1.31
 $ Nxhat2              : num 2

# # save model input data
save(list = c("shrub_gradient_jags.BetNan.data",
              "shrub_gradient_jags.CasTet.data",
              "shrub_gradient_jags.EmpNig.data",
              "shrub_gradient_jags.PhyCae.data",
              "shrub_gradient_jags.RhoGro.data",
              "shrub_gradient_jags.RhoTom.data",
              "shrub_gradient_jags.SalArc.data",
              "shrub_gradient_jags.SalGla.data",
              "shrub_gradient_jags.VacUli.data"),
     file = file.path("..", "data", "processed", "model_input_data_tcws", "shrub_gradient_jags.speciesdata.Rdata"))

> specifying model

write("
  
  model{
    
    # priors
      
      intercept ~ dnorm(0, 0.0001)
      
      b.compet ~ dnorm(0, 0.0001)
      b.shrub_cov ~ dnorm(0, 0.0001)
      b.gramin_cov ~ dnorm(0, 0.0001)
      b.sri ~ dnorm(0, 0.0001)
      b.tri ~ dnorm(0, 0.0001)
      b.tcws ~ dnorm(0, 0.0001)

      sigma.plotgroup ~ dunif(0,100)
      tau.plotgroup <- 1/(sigma.plotgroup * sigma.plotgroup)
      
      b.tempjja.x ~ dnorm(0, 0.001)
      b.tempjja.x2 ~ dnorm(0, 0.001)
      b.tempcont.x ~ dnorm(0, 0.001)
      b.tempcont.x2 ~ dnorm(0, 0.001)
      b.precipjja.x ~ dnorm(0, 0.001)
      b.precipjja.x2 ~ dnorm(0, 0.001)
      
      b.tempXtcws ~ dnorm(0, 0.001)
      b.tempXtcws2 ~ dnorm(0, 0.001)
      
      phi ~ dgamma(0.1, 0.1)
      
      
    # LIKELIHOOD for discrete part

      for (i in 1:N_discrete){ 
        cov.dis[i] ~ dbern(mu[i])
        logit(mu[i]) <- b_plotgroup[plotgroup.dis[i]] + # ~= random effect of plot group
                        b.compet * compet.dis[i] + 
                        b.shrub_cov * shrub_cov.dis[i] + 
                        b.gramin_cov * gramin_cov.dis[i] + 
                        b.tempXtcws * tempjja.dis[i] * tcws.dis[i] +       # for interaction
                        b.tempXtcws2 * (tempjja.dis[i]^2) * tcws.dis[i] +  # for interaction
                        b.tcws * tcws.dis[i] + 
                        b.sri * sri.dis[i] +
                        b.tri * tri.dis[i]
      }
      
      
    # LIKELIHOOD for continuous part

      for (j in 1:N_cont){
        cov.cont[j] ~ dbeta(p[j], q[j])
        p[j] <- mu2[j] * phi
        q[j] <- (1 - mu2[j]) * phi
        logit(mu2[j]) <- b_plotgroup[plotgroup.cont[j]] + # ~= random effect of plot group
                        b.compet * compet.cont[j] +
                        b.shrub_cov * shrub_cov.cont[j] +
                        b.gramin_cov * gramin_cov.cont[j] +
                        b.tempXtcws * tempjja.cont[j] * tcws.cont[j] +       # for interaction
                        b.tempXtcws2 * (tempjja.cont[j]^2) * tcws.cont[j] +  # for interaction
                        b.tcws * tcws.cont[j] + 
                        b.sri * sri.cont[j] +
                        b.tri * tri.cont[j]
      }

      for (k in 1:N_plotgroups){ # length of total plotgroups
        b_plotgroup[k] ~ dnorm(mu.plotgroup[k],tau.plotgroup)
        mu.plotgroup[k] <- intercept + 
                    
                    # plot group level predictors, linear and quadratic term
                    b.tempjja.x * tempjja.tot[k] + 
                    b.tempjja.x2 * (tempjja.tot[k]^2) + 
                    b.tempcont.x * tempcont.tot[k] + 
                    b.tempcont.x2 * (tempcont.tot[k]^2) +
                    b.precipjja.x * precipjja.tot[k] + 
                    b.precipjja.x2 * (precipjja.tot[k]^2) 
      }
      
      
      # add predicted values (derived parameters)
      for (m in 1:Nxhat){
        phat_compet[m] <- intercept + b.compet * xhat_compet[m]
        phat_graminoid_cover[m] <- intercept + b.gramin_cov * xhat_graminoid_cover[m]
        phat_shrub_cover[m] <- intercept + b.shrub_cov * xhat_shrub_cover[m]
        phat_sri[m] <- intercept + b.sri * xhat_sri[m]
        phat_tri[m] <- intercept + b.tri * xhat_tri[m]
        phat_tcws[m] <- intercept + b.tcws * xhat_tcws[m]
        phat_tempjja[m] <- intercept + b.tempjja.x * xhat_tempjja[m] + b.tempjja.x2 * (xhat_tempjja[m]^2)
        phat_tempcont[m] <- intercept + b.tempcont.x * xhat_tempcont[m] + b.tempcont.x2 * (xhat_tempcont[m]^2)
        phat_precipjja[m] <- intercept + b.precipjja.x * xhat_precipjja[m] + b.precipjja.x2 * (xhat_precipjja[m]^2)
      
        for (p in 1:Nxhat2){
          phat_tempXmoist[m,p] <- intercept +
                                      b.tempjja.x * xhat_tempjja[m] +
                                      b.tempjja.x2 * (xhat_tempjja[m]^2) +
                                      b.tcws * xhat_tcws2[p] +
                                      b.tempXtcws * xhat_tempjja[m] * xhat_tcws2[p] +
                                      b.tempXtcws2 * (xhat_tempjja[m]^2) * xhat_tcws2[p]
          }
        }

    
      }
  ", file.path("..", "models_tcws", "shrub_gradient.spec.jags"))

Specify the parameters to be monitored:

params <- c("intercept",
            "b.tempjja.x", "b.tempjja.x2",
            "b.tempcont.x", "b.tempcont.x2",
            "b.precipjja.x", "b.precipjja.x2",
            "b.compet",
            "b.shrub_cov", 
            "b.gramin_cov",
            "b.sri",
            "b.tri",
            "b.tcws",
            "b_plotgroup[1]","b_plotgroup[2]","b_plotgroup[3]","b_plotgroup[63]",
            "sigma.plotgroup",
            "phi",
            "phat_compet", "phat_shrub_cover", "phat_graminoid_cover", 
            "phat_sri", "phat_tcws", "phat_tri", 
            "phat_tempjja", "phat_tempcont", "phat_precipjja", 
            "phat_tempXmoist")

> run & evaluate model

Betula nana

# run model
model_out.shrub_gradient.BetNan <- jags(shrub_gradient_jags.BetNan.data,    # input data
                                        inits = NULL,                       # JAGS to create initial values
                                        params,                             # parameters to be saved
                                        model.file = file.path("..", "models_tcws", "shrub_gradient.spec.jags"), 
                                        n.chains = 3,                       # no. Markov chains
                                        n.iter = 100000, n.burnin = 70000,  # no. iterations & burn-in fraction per chain
                                        n.thin = 2,                         # thinning rate
                                        DIC = FALSE,                        # do not compute deviance, pD, and DIC
                                        working.directory = NULL, 
                                        progress.bar = "text") 

# plot(model_out.shrub_gradient.BetNan) #check convergence, etc.

Extract coefficients and plot effect sizes:

# extract coefficients 
coeff.shrub_gradient.BetNan <- model_out.shrub_gradient.BetNan$BUGSoutput$summary %>% 
  as.data.frame %>% 
  select('mean','sd','2.5%','97.5%','Rhat') %>% 
# add identifying info to data frame
  rownames_to_column(var = "param")
  # mutate(param = as.vector(sapply(strsplit(rownames(coeff.shrub_gradient.BetNan),"[[]",fixed=FALSE), "[", 1))) #%>% print

# add 90% CIs
ci_90.BetNan <- data.frame(q5 = NA, q95 = NA, param = NA)
for (param in 1:(length(model_out.shrub_gradient.BetNan$BUGSoutput$sims.list)-4)){
  ci_90.BetNan[param,1:2] <- quantile(data.frame(model_out.shrub_gradient.BetNan$BUGSoutput$sims.list[param])[,1], probs = c(0.05, 0.95))
  ci_90.BetNan[param, 3] <- names(data.frame(model_out.shrub_gradient.BetNan$BUGSoutput$sims.list))[param]
}

# join to coefficients table
coeff.shrub_gradient.BetNan <- coeff.shrub_gradient.BetNan %>% 
  left_join(ci_90.BetNan, by = "param") %>% 
  # reorder and rename cols
  select(param, mean, sd, 
         l95 = "2.5%",
         l90 = q5,
         u90 = q95,
         u95 = "97.5%",
         Rhat)

save(coeff.shrub_gradient.BetNan, file = file.path("..", "data", "processed", "model_outputs", "species_tcws", "first_runs_converged", "model_output_BetNan.Rdata"))
# load(file.path("..", "data", "processed", "model_outputs", "species_tcws", "first_runs_converged", "model_output_BetNan.Rdata"))
print(coeff.shrub_gradient.BetNan)

(effect_size_plot.BetNan <- model_plot_sig_function(coeff.shrub_gradient.BetNan, title_string = "Betula nana", plot_width = 12.5))

• positive relationship with more conservative communities, negative response to graminoid cover (sig.)
• unimodal relationship with temperature variability (m.s.)

As none of the quadratic terms for climatic variables were significant, they were removed from the model before re-running:

# new model object with terms removed
write("
  
  model{
    
    # priors
      
      intercept ~ dnorm(0, 0.0001)
      
      b.compet ~ dnorm(0, 0.0001)
      b.shrub_cov ~ dnorm(0, 0.0001)
      b.gramin_cov ~ dnorm(0, 0.0001)
      b.sri ~ dnorm(0, 0.0001)
      b.tri ~ dnorm(0, 0.0001)
      b.tcws ~ dnorm(0, 0.0001)

      sigma.plotgroup ~ dunif(0,100)
      tau.plotgroup <- 1/(sigma.plotgroup * sigma.plotgroup)
      
      b.tempjja.x ~ dnorm(0, 0.001)
      # b.tempjja.x2 ~ dnorm(0, 0.001)
      b.tempcont.x ~ dnorm(0, 0.001)
      # b.tempcont.x2 ~ dnorm(0, 0.001)
      b.precipjja.x ~ dnorm(0, 0.001)
      # b.precipjja.x2 ~ dnorm(0, 0.001)
      
      b.tempXtcws ~ dnorm(0, 0.001)
      # b.tempXtcws2 ~ dnorm(0, 0.001)
      
      phi ~ dgamma(0.1, 0.1)
      
      
    # LIKELIHOOD for discrete part

      for (i in 1:N_discrete){ 
        cov.dis[i] ~ dbern(mu[i])
        logit(mu[i]) <- b_plotgroup[plotgroup.dis[i]] + # ~= random effect of plot group
                        b.compet * compet.dis[i] + 
                        b.shrub_cov * shrub_cov.dis[i] + 
                        b.gramin_cov * gramin_cov.dis[i] + 
                        b.tempXtcws * tempjja.dis[i] * tcws.dis[i] +       # for interaction
                        # b.tempXtcws2 * (tempjja.dis[i]^2) * tcws.dis[i] +  # for interaction
                        b.tcws * tcws.dis[i] + 
                        b.sri * sri.dis[i] +
                        b.tri * tri.dis[i]
      }
      
      
    # LIKELIHOOD for continuous part

      for (j in 1:N_cont){
        cov.cont[j] ~ dbeta(p[j], q[j])
        p[j] <- mu2[j] * phi
        q[j] <- (1 - mu2[j]) * phi
        logit(mu2[j]) <- b_plotgroup[plotgroup.cont[j]] + # ~= random effect of plot group
                        b.compet * compet.cont[j] +
                        b.shrub_cov * shrub_cov.cont[j] +
                        b.gramin_cov * gramin_cov.cont[j] +
                        b.tempXtcws * tempjja.cont[j] * tcws.cont[j] +       # for interaction
                        # b.tempXtcws2 * (tempjja.cont[j]^2) * tcws.cont[j] +  # for interaction
                        b.tcws * tcws.cont[j] + 
                        b.sri * sri.cont[j] +
                        b.tri * tri.cont[j]
      }

      for (k in 1:N_plotgroups){ # length of total plotgroups
        b_plotgroup[k] ~ dnorm(mu.plotgroup[k],tau.plotgroup)
        mu.plotgroup[k] <- intercept + 
                    
                    # plot group level predictors, linear and quadratic term
                    b.tempjja.x * tempjja.tot[k] + 
                    # b.tempjja.x2 * (tempjja.tot[k]^2) + 
                    b.tempcont.x * tempcont.tot[k] + 
                    # b.tempcont.x2 * (tempcont.tot[k]^2) +
                    b.precipjja.x * precipjja.tot[k] # + 
                    # b.precipjja.x2 * (precipjja.tot[k]^2) 
      }
      
      
      # add predicted values (derived parameters)
      for (m in 1:Nxhat){
        phat_compet[m] <- intercept + b.compet * xhat_compet[m]
        phat_graminoid_cover[m] <- intercept + b.gramin_cov * xhat_graminoid_cover[m]
        phat_shrub_cover[m] <- intercept + b.shrub_cov * xhat_shrub_cover[m]
        phat_sri[m] <- intercept + b.sri * xhat_sri[m]
        phat_tri[m] <- intercept + b.tri * xhat_tri[m]
        phat_tcws[m] <- intercept + b.tcws * xhat_tcws[m]
        phat_tempjja[m] <- intercept + b.tempjja.x * xhat_tempjja[m] # + b.tempjja.x2 * (xhat_tempjja[m]^2)
        phat_tempcont[m] <- intercept + b.tempcont.x * xhat_tempcont[m] # + b.tempcont.x2 * (xhat_tempcont[m]^2)
        phat_precipjja[m] <- intercept + b.precipjja.x * xhat_precipjja[m] # + b.precipjja.x2 * (xhat_precipjja[m]^2)
      
        for (p in 1:Nxhat2){
          phat_tempXmoist[m,p] <- intercept +
                                      b.tempjja.x * xhat_tempjja[m] +
                                      # b.tempjja.x2 * (xhat_tempjja[m]^2) +
                                      b.tcws * xhat_tcws2[p] +
                                      b.tempXtcws * xhat_tempjja[m] * xhat_tcws2[p] # +
                                      # b.tempXtcws2 * (xhat_tempjja[m]^2) * xhat_tcws2[p]
          }
        }

    
      }
  ", file.path("..", "models_tcws", "shrub_gradient.BetNan2.jags"))

# specify new set of parameters to be monitored
params_BetNan2 <- c("intercept",
                    "b.tempjja.x", # "b.tempjja.x2",
                    "b.tempcont.x", # "b.tempcont.x2",
                    "b.precipjja.x", # "b.precipjja.x2",
                    "b.compet", 
                    "b.shrub_cov",
                    "b.gramin_cov",
                    "b.sri",
                    "b.tri",
                    "b.tcws",
                    "b_plotgroup[1]","b_plotgroup[2]","b_plotgroup[3]","b_plotgroup[63]",
                    "sigma.plotgroup",
                    "phi",
                    "phat_compet", "phat_shrub_cover", "phat_graminoid_cover", 
                    "phat_sri", "phat_tri", "phat_tcws", 
                    "phat_tempjja", "phat_tempcont", "phat_precipjja",
                    "phat_tempXmoist")

model_out.shrub_gradient.BetNan2 <- jags(shrub_gradient_jags.BetNan.data,   # input data
                                        inits = NULL,                       # JAGS to create initial values
                                        params_BetNan2,                     # parameters to be saved
                                        model.file = file.path("..", "models_tcws", "shrub_gradient.BetNan2.jags"), 
                                        n.chains = 3,                       # no. Markov chains
                                        n.iter = 100000, n.burnin = 70000,  # no. iterations & burn-in fraction per chain
                                        n.thin = 2,                         # thinning rate
                                        DIC = FALSE,                        # do not compute deviance, pD, and DIC
                                        working.directory = NULL, 
                                        progress.bar = "text")

# plot(model_out.shrub_gradient.BetNan2) #check convergence, etc.

Extract coefficients and plot effect sizes:

# extract coefficients 
coeff.shrub_gradient.BetNan2 <- model_out.shrub_gradient.BetNan2$BUGSoutput$summary %>% 
  as.data.frame %>% 
  select('mean','sd','2.5%','97.5%','Rhat') %>% 
# add identifying info to data frame
  rownames_to_column(var = "param")

# add 90% CIs
ci_90.BetNan2 <- data.frame(q5 = NA, q95 = NA, param = NA)
for (param in 1:(length(model_out.shrub_gradient.BetNan2$BUGSoutput$sims.list)-4)){
  ci_90.BetNan2[param,1:2] <- quantile(data.frame(model_out.shrub_gradient.BetNan2$BUGSoutput$sims.list[param])[,1], probs = c(0.05, 0.95))
  ci_90.BetNan2[param, 3] <- names(data.frame(model_out.shrub_gradient.BetNan2$BUGSoutput$sims.list))[param]
}

# join to coefficients table
coeff.shrub_gradient.BetNan2 <- coeff.shrub_gradient.BetNan2 %>% 
  left_join(ci_90.BetNan2, by = "param") %>% 
  # reorder and rename cols
  select(param, mean, sd, 
         l95 = "2.5%",
         l90 = q5,
         u90 = q95,
         u95 = "97.5%",
         Rhat) %>% print

save(coeff.shrub_gradient.BetNan2, file = file.path("..", "data", "processed", "model_outputs", "species_tcws", "model_output_BetNan2.Rdata"))
# load(file.path("..", "data", "processed", "model_outputs", "species_tcws", "model_output_BetNan2.Rdata"))

# effect size plot
(effect_size_plot.BetNan2 <- model_plot_sig_function(coeff.shrub_gradient.BetNan2, title_string = "Betula nana", plot_width = 9.5))

• positive response to temperature variability and more conservative communities, negative response to graminoid cover (sig.)
• slightly positive relationship with topographic wetness (n.s.)

Cassiope tetragona

model_out.shrub_gradient.CasTet <- jags(shrub_gradient_jags.CasTet.data,    # input data
                                        inits = NULL,                       # JAGS to create initial values
                                        params,                             # parameters to be saved
                                        model.file = file.path("..", "models_tcws", "shrub_gradient.spec.jags"), 
                                        n.chains = 3,                       # no. Markov chains
                                        n.iter = 100000, n.burnin = 70000,  # no. iterations & burn-in fraction per chain
                                        n.thin = 2,                         # thinning rate
                                        DIC = FALSE,                        # do not compute deviance, pD, and DIC
                                        working.directory = NULL, 
                                        progress.bar = "text")

# plot(model_out.shrub_gradient.CasTet) #check convergence, etc.

Extract coefficients and plot effect sizes:

# extract coefficients 
coeff.shrub_gradient.CasTet <- model_out.shrub_gradient.CasTet$BUGSoutput$summary %>% 
  as.data.frame %>% 
  select('mean','sd','2.5%','97.5%','Rhat') %>% 
# add identifying info to data frame
  rownames_to_column(var = "param")
  # mutate(param = as.vector(sapply(strsplit(rownames(coeff.shrub_gradient.CasTet),"[[]",fixed=FALSE), "[", 1))) #%>% print

# add 90% CIs
ci_90.CasTet <- data.frame(q5 = NA, q95 = NA, param = NA)
for (param in 1:(length(model_out.shrub_gradient.CasTet$BUGSoutput$sims.list)-4)){
  ci_90.CasTet[param,1:2] <- quantile(data.frame(model_out.shrub_gradient.CasTet$BUGSoutput$sims.list[param])[,1], probs = c(0.05, 0.95))
  ci_90.CasTet[param, 3] <- names(data.frame(model_out.shrub_gradient.CasTet$BUGSoutput$sims.list))[param]
}

# join to coefficients table
coeff.shrub_gradient.CasTet <- coeff.shrub_gradient.CasTet %>% 
  left_join(ci_90.CasTet, by = "param") %>% 
  # reorder and rename cols
  select(param, mean, sd, 
         l95 = "2.5%",
         l90 = q5,
         u90 = q95,
         u95 = "97.5%",
         Rhat)

print(coeff.shrub_gradient.CasTet)

save(coeff.shrub_gradient.CasTet, file = file.path("..", "data", "processed", "model_outputs", "species_tcws", "model_output_CasTet.Rdata"))

• not converging: large R-hat values (~ 1.3) (only 16 non-zero values across plot groups)

Empetrum nigrum

model_out.shrub_gradient.EmpNig <- jags(shrub_gradient_jags.EmpNig.data,    # input data
                                        inits = NULL,                       # JAGS to create initial values
                                        params,                             # parameters to be saved
                                        model.file = file.path("..", "models_tcws", "shrub_gradient.spec.jags"), 
                                        n.chains = 3,                       # no. Markov chains
                                        n.iter = 100000, n.burnin = 70000,  # no. iterations & burn-in fraction per chain
                                        n.thin = 2,                         # thinning rate
                                        DIC = FALSE,                        # do not compute deviance, pD, and DIC
                                        working.directory = NULL, 
                                        progress.bar = "text")

# plot(model_out.shrub_gradient.EmpNig) #check convergence, etc.

Extract coefficients and plot effect sizes:

# extract coefficients 
coeff.shrub_gradient.EmpNig <- model_out.shrub_gradient.EmpNig$BUGSoutput$summary %>% 
  as.data.frame %>% 
  select('mean','sd','2.5%','97.5%','Rhat') %>% 
# add identifying info to data frame
  rownames_to_column(var = "param")
  # mutate(param = as.vector(sapply(strsplit(rownames(coeff.shrub_gradient.EmpNig),"[[]",fixed=FALSE), "[", 1))) #%>% print

# add 90% CIs
ci_90.EmpNig <- data.frame(q5 = NA, q95 = NA, param = NA)
for (param in 1:(length(model_out.shrub_gradient.EmpNig$BUGSoutput$sims.list)-4)){
  ci_90.EmpNig[param,1:2] <- quantile(data.frame(model_out.shrub_gradient.EmpNig$BUGSoutput$sims.list[param])[,1], probs = c(0.05, 0.95))
  ci_90.EmpNig[param, 3] <- names(data.frame(model_out.shrub_gradient.EmpNig$BUGSoutput$sims.list))[param]
}

# join to coefficients table
coeff.shrub_gradient.EmpNig <- coeff.shrub_gradient.EmpNig %>% 
  left_join(ci_90.EmpNig, by = "param") %>% 
  # reorder and rename cols
  select(param, mean, sd, 
         l95 = "2.5%",
         l90 = q5,
         u90 = q95,
         u95 = "97.5%",
         Rhat) %>% print

save(coeff.shrub_gradient.EmpNig, file = file.path("..", "data", "processed", "model_outputs", "species_tcws", "first_runs_converged", "model_output_EmpNig.Rdata"))
# load(file.path("..", "data", "processed", "model_outputs", "species_tcws", "first_runs_converged", "model_output_EmpNig.Rdata"))

# effect size plot
(effect_size_plot.EmpNig <- model_plot_marg_function(coeff.shrub_gradient.EmpNig, title_string = "Empetrum nigrum", plot_width = 12.5))

• positive linear relationship to summer precipitation and TC wetness (sig.)

As none of the quadratic terms for climatic variables were significant, they were removed from the model before re-running:

# new model object with terms removed
write("
  
  model{
    
    # priors
      
      intercept ~ dnorm(0, 0.0001)
      
      b.compet ~ dnorm(0, 0.0001)
      b.shrub_cov ~ dnorm(0, 0.0001)
      b.gramin_cov ~ dnorm(0, 0.0001)
      b.sri ~ dnorm(0, 0.0001)
      b.tri ~ dnorm(0, 0.0001)
      b.tcws ~ dnorm(0, 0.0001)

      sigma.plotgroup ~ dunif(0,100)
      tau.plotgroup <- 1/(sigma.plotgroup * sigma.plotgroup)
      
      b.tempjja.x ~ dnorm(0, 0.001)
      # b.tempjja.x2 ~ dnorm(0, 0.001)
      b.tempcont.x ~ dnorm(0, 0.001)
      # b.tempcont.x2 ~ dnorm(0, 0.001)
      b.precipjja.x ~ dnorm(0, 0.001)
      # b.precipjja.x2 ~ dnorm(0, 0.001)
      
      b.tempXtcws ~ dnorm(0, 0.001)
      # b.tempXtcws2 ~ dnorm(0, 0.001)
      
      phi ~ dgamma(0.1, 0.1)
      
      
    # LIKELIHOOD for discrete part

      for (i in 1:N_discrete){ 
        cov.dis[i] ~ dbern(mu[i])
        logit(mu[i]) <- b_plotgroup[plotgroup.dis[i]] + # ~= random effect of plot group
                        b.compet * compet.dis[i] + 
                        b.shrub_cov * shrub_cov.dis[i] + 
                        b.gramin_cov * gramin_cov.dis[i] + 
                        b.tempXtcws * tempjja.dis[i] * tcws.dis[i] +       # for interaction
                        # b.tempXtcws2 * (tempjja.dis[i]^2) * tcws.dis[i] +  # for interaction
                        b.tcws * tcws.dis[i] + 
                        b.sri * sri.dis[i] +
                        b.tri * tri.dis[i]
      }
      
      
    # LIKELIHOOD for continuous part

      for (j in 1:N_cont){
        cov.cont[j] ~ dbeta(p[j], q[j])
        p[j] <- mu2[j] * phi
        q[j] <- (1 - mu2[j]) * phi
        logit(mu2[j]) <- b_plotgroup[plotgroup.cont[j]] + # ~= random effect of plot group
                        b.compet * compet.cont[j] +
                        b.shrub_cov * shrub_cov.cont[j] +
                        b.gramin_cov * gramin_cov.cont[j] +
                        b.tempXtcws * tempjja.cont[j] * tcws.cont[j] +       # for interaction
                        # b.tempXtcws2 * (tempjja.cont[j]^2) * tcws.cont[j] +  # for interaction
                        b.tcws * tcws.cont[j] + 
                        b.sri * sri.cont[j] +
                        b.tri * tri.cont[j]
      }

      for (k in 1:N_plotgroups){ # length of total plotgroups
        b_plotgroup[k] ~ dnorm(mu.plotgroup[k],tau.plotgroup)
        mu.plotgroup[k] <- intercept + 
                    
                    # plot group level predictors, linear and quadratic term
                    b.tempjja.x * tempjja.tot[k] + 
                    # b.tempjja.x2 * (tempjja.tot[k]^2) + 
                    b.tempcont.x * tempcont.tot[k] + 
                    # b.tempcont.x2 * (tempcont.tot[k]^2) +
                    b.precipjja.x * precipjja.tot[k] # + 
                    # b.precipjja.x2 * (precipjja.tot[k]^2) 
      }
      
      
      # add predicted values (derived parameters)
      for (m in 1:Nxhat){
        phat_compet[m] <- intercept + b.compet * xhat_compet[m]
        phat_graminoid_cover[m] <- intercept + b.gramin_cov * xhat_graminoid_cover[m]
        phat_shrub_cover[m] <- intercept + b.shrub_cov * xhat_shrub_cover[m]
        phat_sri[m] <- intercept + b.sri * xhat_sri[m]
        phat_tri[m] <- intercept + b.tri * xhat_tri[m]
        phat_tcws[m] <- intercept + b.tcws * xhat_tcws[m]
        phat_tempjja[m] <- intercept + b.tempjja.x * xhat_tempjja[m] # + b.tempjja.x2 * (xhat_tempjja[m]^2)
        phat_tempcont[m] <- intercept + b.tempcont.x * xhat_tempcont[m] # + b.tempcont.x2 * (xhat_tempcont[m]^2)
        phat_precipjja[m] <- intercept + b.precipjja.x * xhat_precipjja[m] # + b.precipjja.x2 * (xhat_precipjja[m]^2)
      
        for (p in 1:Nxhat2){
          phat_tempXmoist[m,p] <- intercept +
                                      b.tempjja.x * xhat_tempjja[m] +
                                      # b.tempjja.x2 * (xhat_tempjja[m]^2) +
                                      b.tcws * xhat_tcws2[p] +
                                      b.tempXtcws * xhat_tempjja[m] * xhat_tcws2[p] # +
                                      # b.tempXtcws2 * (xhat_tempjja[m]^2) * xhat_tcws2[p]
          }
        }

    
      }
  ", file.path("..", "models_tcws", "shrub_gradient.EmpNig2.jags"))

# specify new set of parameters to be monitored
params_EmpNig2 <- c("intercept",
                    "b.tempjja.x", # "b.tempjja.x2",
                    "b.tempcont.x", # "b.tempcont.x2",
                    "b.precipjja.x", # "b.precipjja.x2",
                    "b.compet", 
                    "b.shrub_cov",
                    "b.gramin_cov",
                    "b.sri",
                    "b.tri",
                    "b.tcws",
                    "b_plotgroup[1]","b_plotgroup[2]","b_plotgroup[3]","b_plotgroup[63]",
                    "sigma.plotgroup",
                    "phi",
                    "phat_compet", "phat_shrub_cover", "phat_graminoid_cover", 
                    "phat_sri", "phat_tri", "phat_tcws", 
                    "phat_tempjja", "phat_tempcont", "phat_precipjja",
                    "phat_tempXmoist")

model_out.shrub_gradient.EmpNig2 <- jags(shrub_gradient_jags.EmpNig.data,    # input data
                                        inits = NULL,                       # JAGS to create initial values
                                        params_EmpNig2,                             # parameters to be saved
                                        model.file = file.path("..", "models_tcws", "shrub_gradient.EmpNig2.jags"), 
                                        n.chains = 3,                       # no. Markov chains
                                        n.iter = 100000, n.burnin = 70000,  # no. iterations & burn-in fraction per chain
                                        n.thin = 2,                         # thinning rate
                                        DIC = FALSE,                        # do not compute deviance, pD, and DIC
                                        working.directory = NULL, 
                                        progress.bar = "text")

# plot(model_out.shrub_gradient.EmpNig2) #check convergence, etc.

Extract coefficients and plot effect sizes:

# extract coefficients 
coeff.shrub_gradient.EmpNig2 <- model_out.shrub_gradient.EmpNig2$BUGSoutput$summary %>% 
  as.data.frame %>% 
  select('mean','sd','2.5%','97.5%','Rhat') %>% 
# add identifying info to data frame
  rownames_to_column(var = "param")

# add 90% CIs
ci_90.EmpNig2 <- data.frame(q5 = NA, q95 = NA, param = NA)
for (param in 1:(length(model_out.shrub_gradient.EmpNig2$BUGSoutput$sims.list)-4)){
  ci_90.EmpNig2[param,1:2] <- quantile(data.frame(model_out.shrub_gradient.EmpNig2$BUGSoutput$sims.list[param])[,1], probs = c(0.05, 0.95))
  ci_90.EmpNig2[param, 3] <- names(data.frame(model_out.shrub_gradient.EmpNig2$BUGSoutput$sims.list))[param]
}

# join to coefficients table
coeff.shrub_gradient.EmpNig2 <- coeff.shrub_gradient.EmpNig2 %>% 
  left_join(ci_90.EmpNig2, by = "param") %>% 
  # reorder and rename cols
  select(param, mean, sd, 
         l95 = "2.5%",
         l90 = q5,
         u90 = q95,
         u95 = "97.5%",
         Rhat) %>% print

save(coeff.shrub_gradient.EmpNig2, file = file.path("..", "data", "processed", "model_outputs", "species_tcws", "model_output_EmpNig2.Rdata"))
# load(file.path("..", "data", "processed", "model_outputs", "species_tcws", "model_output_EmpNig2.Rdata"))

# effect size plot
(effect_size_plot.EmpNig2 <- model_plot_sig_function(coeff.shrub_gradient.EmpNig2, title_string = "Empetrum nigrum", plot_width = 9.5))

• negative response to temperature variability, positive relationship with precipitation (linear) and TCwetness (sig.)
• slightly negative response to graminoid cover (n.s.)

Phyllodoce caerulea

model_out.shrub_gradient.PhyCae <- jags(shrub_gradient_jags.PhyCae.data,    # input data
                                        inits = NULL,                       # JAGS to create initial values
                                        params,                             # parameters to be saved
                                        model.file = file.path("..", "models_tcws", "shrub_gradient.spec.jags"), 
                                        n.chains = 3,                       # no. Markov chains
                                        n.iter = 100000, n.burnin = 70000,  # no. iterations & burn-in fraction per chain
                                        n.thin = 2,                         # thinning rate
                                        DIC = FALSE,                        # do not compute deviance, pD, and DIC
                                        working.directory = NULL, 
                                        progress.bar = "text")

# plot(model_out.shrub_gradient.PhyCae) #check convergence, etc.

Extract coefficients and plot effect sizes:

# extract coefficients 
coeff.shrub_gradient.PhyCae <- model_out.shrub_gradient.PhyCae$BUGSoutput$summary %>% 
  as.data.frame %>% 
  select('mean','sd','2.5%','97.5%','Rhat') %>% 
# add identifying info to data frame
  rownames_to_column(var = "param")
  # mutate(param = as.vector(sapply(strsplit(rownames(coeff.shrub_gradient.PhyCae),"[[]",fixed=FALSE), "[", 1))) #%>% print

# add 90% CIs
ci_90.PhyCae <- data.frame(q5 = NA, q95 = NA, param = NA)
for (param in 1:(length(model_out.shrub_gradient.PhyCae$BUGSoutput$sims.list)-4)){
  ci_90.PhyCae[param,1:2] <- quantile(data.frame(model_out.shrub_gradient.PhyCae$BUGSoutput$sims.list[param])[,1], probs = c(0.05, 0.95))
  ci_90.PhyCae[param, 3] <- names(data.frame(model_out.shrub_gradient.PhyCae$BUGSoutput$sims.list))[param]
}

# join to coefficients table
coeff.shrub_gradient.PhyCae <- coeff.shrub_gradient.PhyCae %>% 
  left_join(ci_90.PhyCae, by = "param") %>% 
  # reorder and rename cols
  select(param, mean, sd, 
         l95 = "2.5%",
         l90 = q5,
         u90 = q95,
         u95 = "97.5%",
         Rhat) %>% print

save(coeff.shrub_gradient.PhyCae, file = file.path("..", "data", "processed", "model_outputs", "species_tcws", "model_output_PhyCae.Rdata"))

• not converged well relatively large R-hat values (~ 1.02) (only 10 non-zero values across plot groups)

Rhododendron groenlandicum

model_out.shrub_gradient.RhoGro <- jags(shrub_gradient_jags.RhoGro.data,    # input data
                                        inits = NULL,                       # JAGS to create initial values
                                        params,                             # parameters to be saved
                                        model.file = file.path("..", "models_tcws", "shrub_gradient.spec.jags"), 
                                        n.chains = 3,                       # no. Markov chains
                                        n.iter = 100000, n.burnin = 70000,  # no. iterations & burn-in fraction per chain
                                        n.thin = 2,                         # thinning rate
                                        DIC = FALSE,                        # do not compute deviance, pD, and DIC
                                        working.directory = NULL, 
                                        progress.bar = "text")

# plot(model_out.shrub_gradient.RhoGro) #check convergence, etc.

Extract coefficients and plot effect sizes:

# extract coefficients 
coeff.shrub_gradient.RhoGro <- model_out.shrub_gradient.RhoGro$BUGSoutput$summary %>% 
  as.data.frame %>% 
  select('mean','sd','2.5%','97.5%','Rhat') %>% 
# add identifying info to data frame
  rownames_to_column(var = "param")
  # mutate(param = as.vector(sapply(strsplit(rownames(coeff.shrub_gradient.RhoGro),"[[]",fixed=FALSE), "[", 1))) #%>% print

# add 90% CIs
ci_90.RhoGro <- data.frame(q5 = NA, q95 = NA, param = NA)
for (param in 1:(length(model_out.shrub_gradient.RhoGro$BUGSoutput$sims.list)-4)){
  ci_90.RhoGro[param,1:2] <- quantile(data.frame(model_out.shrub_gradient.RhoGro$BUGSoutput$sims.list[param])[,1], probs = c(0.05, 0.95))
  ci_90.RhoGro[param, 3] <- names(data.frame(model_out.shrub_gradient.RhoGro$BUGSoutput$sims.list))[param]
}

# join to coefficients table
coeff.shrub_gradient.RhoGro <- coeff.shrub_gradient.RhoGro %>% 
  left_join(ci_90.RhoGro, by = "param") %>% 
  # reorder and rename cols
  select(param, mean, sd, 
         l95 = "2.5%",
         l90 = q5,
         u90 = q95,
         u95 = "97.5%",
         Rhat) %>% print

save(coeff.shrub_gradient.RhoGro, file = file.path("..", "data", "processed", "model_outputs", "species_tcws", "first_runs_converged", "model_output_RhoGro.Rdata"))
# load(file.path("..", "data", "processed", "model_outputs", "species_tcws", "first_runs_converged", "model_output_RhoGro.Rdata"))

# effect size plot
(effect_size_plot.RhoGro <- model_plot_marg_function(coeff.shrub_gradient.RhoGro, title_string = "Rhododendron groenlandicum", plot_width = 12.5))

• negative response to graminoid cover (m.s.)
• slightly positive response to temperature variability, solar radiation, terrain heterogeneity, TC wetness; slightly unimodal response to summer temperature (n.s.)

As the quadratic terms for climatic variables were not significant, they were removed from the model before re-running:

# new model object with terms removed
write("
  
  model{
    
    # priors
      
      intercept ~ dnorm(0, 0.0001)
      
      b.compet ~ dnorm(0, 0.0001)
      b.shrub_cov ~ dnorm(0, 0.0001)
      b.gramin_cov ~ dnorm(0, 0.0001)
      b.sri ~ dnorm(0, 0.0001)
      b.tri ~ dnorm(0, 0.0001)
      b.tcws ~ dnorm(0, 0.0001)

      sigma.plotgroup ~ dunif(0,100)
      tau.plotgroup <- 1/(sigma.plotgroup * sigma.plotgroup)
      
      b.tempjja.x ~ dnorm(0, 0.001)
      # b.tempjja.x2 ~ dnorm(0, 0.001)
      b.tempcont.x ~ dnorm(0, 0.001)
      # b.tempcont.x2 ~ dnorm(0, 0.001)
      b.precipjja.x ~ dnorm(0, 0.001)
      # b.precipjja.x2 ~ dnorm(0, 0.001)
      
      b.tempXtcws ~ dnorm(0, 0.001)
      b.tempXtcws2 ~ dnorm(0, 0.001)
      
      phi ~ dgamma(0.1, 0.1)
      
      
    # LIKELIHOOD for discrete part

      for (i in 1:N_discrete){ 
        cov.dis[i] ~ dbern(mu[i])
        logit(mu[i]) <- b_plotgroup[plotgroup.dis[i]] + # ~= random effect of plot group
                        b.compet * compet.dis[i] + 
                        b.shrub_cov * shrub_cov.dis[i] + 
                        b.gramin_cov * gramin_cov.dis[i] + 
                        b.tempXtcws * tempjja.dis[i] * tcws.dis[i] +       # for interaction
                        # b.tempXtcws2 * (tempjja.dis[i]^2) * tcws.dis[i] +  # for interaction
                        b.tcws * tcws.dis[i] + 
                        b.sri * sri.dis[i] +
                        b.tri * tri.dis[i]
      }
      
      
    # LIKELIHOOD for continuous part

      for (j in 1:N_cont){
        cov.cont[j] ~ dbeta(p[j], q[j])
        p[j] <- mu2[j] * phi
        q[j] <- (1 - mu2[j]) * phi
        logit(mu2[j]) <- b_plotgroup[plotgroup.cont[j]] + # ~= random effect of plot group
                        b.compet * compet.cont[j] +
                        b.shrub_cov * shrub_cov.cont[j] +
                        b.gramin_cov * gramin_cov.cont[j] +
                        b.tempXtcws * tempjja.cont[j] * tcws.cont[j] +       # for interaction
                        # b.tempXtcws2 * (tempjja.cont[j]^2) * tcws.cont[j] +  # for interaction
                        b.tcws * tcws.cont[j] + 
                        b.sri * sri.cont[j] +
                        b.tri * tri.cont[j]
      }

      for (k in 1:N_plotgroups){ # length of total plotgroups
        b_plotgroup[k] ~ dnorm(mu.plotgroup[k],tau.plotgroup)
        mu.plotgroup[k] <- intercept + 
                    
                    # plot group level predictors, linear and quadratic term
                    b.tempjja.x * tempjja.tot[k] + 
                    # b.tempjja.x2 * (tempjja.tot[k]^2) + 
                    b.tempcont.x * tempcont.tot[k] + 
                    # b.tempcont.x2 * (tempcont.tot[k]^2) +
                    b.precipjja.x * precipjja.tot[k] # + 
                    # b.precipjja.x2 * (precipjja.tot[k]^2) 
      }
      
      
      # add predicted values (derived parameters)
      for (m in 1:Nxhat){
        phat_compet[m] <- intercept + b.compet * xhat_compet[m]
        phat_graminoid_cover[m] <- intercept + b.gramin_cov * xhat_graminoid_cover[m]
        phat_shrub_cover[m] <- intercept + b.shrub_cov * xhat_shrub_cover[m]
        phat_sri[m] <- intercept + b.sri * xhat_sri[m]
        phat_tri[m] <- intercept + b.tri * xhat_tri[m]
        phat_tcws[m] <- intercept + b.tcws * xhat_tcws[m]
        phat_tempjja[m] <- intercept + b.tempjja.x * xhat_tempjja[m] # + b.tempjja.x2 * (xhat_tempjja[m]^2)
        phat_tempcont[m] <- intercept + b.tempcont.x * xhat_tempcont[m] # + b.tempcont.x2 * (xhat_tempcont[m]^2)
        phat_precipjja[m] <- intercept + b.precipjja.x * xhat_precipjja[m] # + b.precipjja.x2 * (xhat_precipjja[m]^2)
      
        for (p in 1:Nxhat2){
          phat_tempXmoist[m,p] <- intercept +
                                      b.tempjja.x * xhat_tempjja[m] +
                                      # b.tempjja.x2 * (xhat_tempjja[m]^2) +
                                      b.tcws * xhat_tcws2[p] +
                                      b.tempXtcws * xhat_tempjja[m] * xhat_tcws2[p] # +
                                      # b.tempXtcws2 * (xhat_tempjja[m]^2) * xhat_tcws2[p]
          }
        }

    
      }
  ", file.path("..", "models_tcws", "shrub_gradient.RhoGro2.jags"))

# specify new set of parameters to be monitored
params_RhoGro2 <- c("intercept",
                    "b.tempjja.x", # "b.tempjja.x2",
                    "b.tempcont.x", # "b.tempcont.x2",
                    "b.precipjja.x", # "b.precipjja.x2",
                    "b.compet", 
                    "b.shrub_cov",
                    "b.gramin_cov",
                    "b.sri",
                    "b.tri",
                    "b.tcws",
                    "b_plotgroup[1]","b_plotgroup[2]","b_plotgroup[3]","b_plotgroup[63]",
                    "sigma.plotgroup",
                    "phi",
                    "phat_compet", 
                    "phat_shrub_cover", "phat_graminoid_cover", 
                    "phat_sri", "phat_tri", "phat_tcws", 
                    "phat_tempjja", "phat_tempcont", "phat_precipjja",
                    "phat_tempXmoist")

model_out.shrub_gradient.RhoGro2 <- jags(shrub_gradient_jags.RhoGro.data,    # input data
                                        inits = NULL,                       # JAGS to create initial values
                                        params_RhoGro2,                             # parameters to be saved
                                        model.file = file.path("..", "models_tcws", "shrub_gradient.RhoGro2.jags"), 
                                        n.chains = 3,                       # no. Markov chains
                                        n.iter = 100000, n.burnin = 70000,  # no. iterations & burn-in fraction per chain
                                        n.thin = 2,                         # thinning rate
                                        DIC = FALSE,                        # do not compute deviance, pD, and DIC
                                        working.directory = NULL, 
                                        progress.bar = "text")

# plot(model_out.shrub_gradient.RhoGro2) #check convergence, etc.

Extract coefficients and plot effect sizes:

# extract coefficients 
coeff.shrub_gradient.RhoGro2 <- model_out.shrub_gradient.RhoGro2$BUGSoutput$summary %>% 
  as.data.frame %>% 
  select('mean','sd','2.5%','97.5%','Rhat') %>% 
# add identifying info to data frame
  rownames_to_column(var = "param")
  # mutate(param = as.vector(sapply(strsplit(rownames(coeff.shrub_gradient.RhoGro),"[[]",fixed=FALSE), "[", 1))) #%>% print

# add 90% CIs
ci_90.RhoGro2 <- data.frame(q5 = NA, q95 = NA, param = NA)
for (param in 1:(length(model_out.shrub_gradient.RhoGro2$BUGSoutput$sims.list)-4)){
  ci_90.RhoGro2[param,1:2] <- quantile(data.frame(model_out.shrub_gradient.RhoGro2$BUGSoutput$sims.list[param])[,1], probs = c(0.05, 0.95))
  ci_90.RhoGro2[param, 3] <- names(data.frame(model_out.shrub_gradient.RhoGro2$BUGSoutput$sims.list))[param]
}

# join to coefficients table
coeff.shrub_gradient.RhoGro2 <- coeff.shrub_gradient.RhoGro2 %>% 
  left_join(ci_90.RhoGro2, by = "param") %>% 
  # reorder and rename cols
  select(param, mean, sd, 
         l95 = "2.5%",
         l90 = q5,
         u90 = q95,
         u95 = "97.5%",
         Rhat) %>% print

save(coeff.shrub_gradient.RhoGro2, file = file.path("..", "data", "processed", "model_outputs", "species_tcws", "model_output_RhoGro2.Rdata"))
# load(file.path("..", "data", "processed", "model_outputs", "species_tcws", "model_output_RhoGro2.Rdata"))

# effect size plot
(effect_size_plot.RhoGro2 <- model_plot_marg_function(coeff.shrub_gradient.RhoGro2, title_string = "Rhododendron groenlandicum", plot_width = 9.5))

• positive relationship with TCwetness (sig.), positive relationship with temperature variability, negative with graminoid cover (m.s.)
• negative relationship with precipitation (linear) (n.s.)

Rhododendron tomentosum

model_out.shrub_gradient.RhoTom <- jags(shrub_gradient_jags.RhoTom.data,    # input data
                                        inits = NULL,                       # JAGS to create initial values
                                        params,                             # parameters to be saved
                                        model.file = file.path("..", "models_tcws", "shrub_gradient.spec.jags"), 
                                        n.chains = 3,                       # no. Markov chains
                                        n.iter = 100000, n.burnin = 70000,  # no. iterations & burn-in fraction per chain
                                        n.thin = 2,                         # thinning rate
                                        DIC = FALSE,                        # do not compute deviance, pD, and DIC
                                        working.directory = NULL, 
                                        progress.bar = "text")

# plot(model_out.shrub_gradient.RhoTom) #check convergence, etc.

Extract coefficients and plot effect sizes:

# extract coefficients 
coeff.shrub_gradient.RhoTom <- model_out.shrub_gradient.RhoTom$BUGSoutput$summary %>% 
  as.data.frame %>% 
  select('mean','sd','2.5%','97.5%','Rhat') %>% 
# add identifying info to data frame
  rownames_to_column(var = "param")
  # mutate(param = as.vector(sapply(strsplit(rownames(coeff.shrub_gradient.RhoTom),"[[]",fixed=FALSE), "[", 1))) #%>% print

# add 90% CIs
ci_90.RhoTom <- data.frame(q5 = NA, q95 = NA, param = NA)
for (param in 1:(length(model_out.shrub_gradient.RhoTom$BUGSoutput$sims.list)-4)){
  ci_90.RhoTom[param,1:2] <- quantile(data.frame(model_out.shrub_gradient.RhoTom$BUGSoutput$sims.list[param])[,1], probs = c(0.05, 0.95))
  ci_90.RhoTom[param, 3] <- names(data.frame(model_out.shrub_gradient.RhoTom$BUGSoutput$sims.list))[param]
}

# join to coefficients table
coeff.shrub_gradient.RhoTom <- coeff.shrub_gradient.RhoTom %>% 
  left_join(ci_90.RhoTom, by = "param") %>% 
  # reorder and rename cols
  select(param, mean, sd, 
         l95 = "2.5%",
         l90 = q5,
         u90 = q95,
         u95 = "97.5%",
         Rhat) %>% print

save(coeff.shrub_gradient.RhoTom, file = file.path("..", "data", "processed", "model_outputs", "species_tcws", "model_output_RhoTom.Rdata"))

• not converging: really large R-hat values (~ 2.7), large variation in parameter estimates (only 13 non-zero values)

Salix arctophila

model_out.shrub_gradient.SalArc <- jags(shrub_gradient_jags.SalArc.data,    # input data
                                        inits = NULL,                       # JAGS to create initial values
                                        params,                             # parameters to be saved
                                        model.file = file.path("..", "models_tcws", "shrub_gradient.spec.jags"), 
                                        n.chains = 3,                       # no. Markov chains
                                        n.iter = 100000, n.burnin = 70000,  # no. iterations & burn-in fraction per chain
                                        n.thin = 2,                         # thinning rate
                                        DIC = FALSE,                        # do not compute deviance, pD, and DIC
                                        working.directory = NULL, 
                                        progress.bar = "text")

# plot(model_out.shrub_gradient.SalArc) #check convergence, etc.

Extract coefficients and plot effect sizes:

# extract coefficients 
coeff.shrub_gradient.SalArc <- model_out.shrub_gradient.SalArc$BUGSoutput$summary %>% 
  as.data.frame %>% 
  select('mean','sd','2.5%','97.5%','Rhat') %>% 
# add identifying info to data frame
  rownames_to_column(var = "param")
  # mutate(param = as.vector(sapply(strsplit(rownames(coeff.shrub_gradient.SalArc),"[[]",fixed=FALSE), "[", 1))) #%>% print

# add 90% CIs
ci_90.SalArc <- data.frame(q5 = NA, q95 = NA, param = NA)
for (param in 1:(length(model_out.shrub_gradient.SalArc$BUGSoutput$sims.list)-4)){
  ci_90.SalArc[param,1:2] <- quantile(data.frame(model_out.shrub_gradient.SalArc$BUGSoutput$sims.list[param])[,1], probs = c(0.05, 0.95))
  ci_90.SalArc[param, 3] <- names(data.frame(model_out.shrub_gradient.SalArc$BUGSoutput$sims.list))[param]
}

# join to coefficients table
coeff.shrub_gradient.SalArc <- coeff.shrub_gradient.SalArc %>% 
  left_join(ci_90.SalArc, by = "param") %>% 
  # reorder and rename cols
  select(param, mean, sd, 
         l95 = "2.5%",
         l90 = q5,
         u90 = q95,
         u95 = "97.5%",
         Rhat) %>% print

# R-hats look okay => check traceplots
# traceplot(coeff.shrub_gradient.SalArc)

save(coeff.shrub_gradient.SalArc, file = file.path("..", "data", "processed", "model_outputs", "species_tcws", "model_output_SalArc.Rdata"))

• not converging well: relatively large R-hat values (< 1.04), and quite some variation in trace plots, especially for climatic parameters (only 13 non-zero values across plot groups)

Salix glauca

model_out.shrub_gradient.SalGla <- jags(shrub_gradient_jags.SalGla.data,    # input data
                                        inits = NULL,                       # JAGS to create initial values
                                        params,                             # parameters to be saved
                                        model.file = file.path("..", "models_tcws", "shrub_gradient.spec.jags"), 
                                        n.chains = 3,                       # no. Markov chains
                                        n.iter = 100000, n.burnin = 70000,  # no. iterations & burn-in fraction per chain
                                        n.thin = 2,                         # thinning rate
                                        DIC = FALSE,                        # do not compute deviance, pD, and DIC
                                        working.directory = NULL, 
                                        progress.bar = "text")

# plot(model_out.shrub_gradient.SalGla) #check convergence, etc.

Extract coefficients and plot effect sizes:

# extract coefficients 
coeff.shrub_gradient.SalGla <- model_out.shrub_gradient.SalGla$BUGSoutput$summary %>% 
  as.data.frame %>% 
  select('mean','sd','2.5%','97.5%','Rhat') %>% 
# add identifying info to data frame
  rownames_to_column(var = "param")
  # mutate(param = as.vector(sapply(strsplit(rownames(coeff.shrub_gradient.SalGla),"[[]",fixed=FALSE), "[", 1))) #%>% print

# add 90% CIs
ci_90.SalGla <- data.frame(q5 = NA, q95 = NA, param = NA)
for (param in 1:(length(model_out.shrub_gradient.SalGla$BUGSoutput$sims.list)-4)){
  ci_90.SalGla[param,1:2] <- quantile(data.frame(model_out.shrub_gradient.SalGla$BUGSoutput$sims.list[param])[,1], probs = c(0.05, 0.95))
  ci_90.SalGla[param, 3] <- names(data.frame(model_out.shrub_gradient.SalGla$BUGSoutput$sims.list))[param]
}

# join to coefficients table
coeff.shrub_gradient.SalGla <- coeff.shrub_gradient.SalGla %>% 
  left_join(ci_90.SalGla, by = "param") %>% 
  # reorder and rename cols
  select(param, mean, sd, 
         l95 = "2.5%",
         l90 = q5,
         u90 = q95,
         u95 = "97.5%",
         Rhat) %>% print

save(coeff.shrub_gradient.SalGla, file = file.path("..", "data", "processed", "model_outputs", "species_tcws", "first_runs_converged", "model_output_SalGla.Rdata"))
# load(file.path("..", "data", "processed", "model_outputs", "species_tcws", "first_runs_converged", "model_output_SalGla.Rdata"))

# effect size plot
(effect_size_plot.SalGla <- model_plot_marg_function(coeff.shrub_gradient.SalGla, title_string = "Salix glauca", plot_width = 12.5))

• positive relationship with less acquisitive communities (sig.), other shrub cover and temperature variability (m.s.)
• slightly negative response to summer temperature (n.s.)

As the quadratic terms for summer temperature, temperature variability and summer precipitation were not significant, they were removed from the model before re-running:

# new model object with terms removed
write("
  
  model{

    # priors
      
      intercept ~ dnorm(0, 0.0001)
      
      b.compet ~ dnorm(0, 0.0001)
      b.shrub_cov ~ dnorm(0, 0.0001)
      b.gramin_cov ~ dnorm(0, 0.0001)
      b.sri ~ dnorm(0, 0.0001)
      b.tri ~ dnorm(0, 0.0001)
      b.tcws ~ dnorm(0, 0.0001)

      sigma.plotgroup ~ dunif(0,100)
      tau.plotgroup <- 1/(sigma.plotgroup * sigma.plotgroup)
      
      b.tempjja.x ~ dnorm(0, 0.001)
      # b.tempjja.x2 ~ dnorm(0, 0.001)
      b.tempcont.x ~ dnorm(0, 0.001)
      # b.tempcont.x2 ~ dnorm(0, 0.001)
      b.precipjja.x ~ dnorm(0, 0.001)
      # b.precipjja.x2 ~ dnorm(0, 0.001)
      
      b.tempXtcws ~ dnorm(0, 0.001)
      # b.tempXtcws2 ~ dnorm(0, 0.001)
      
      phi ~ dgamma(0.1, 0.1)
      
      
    # LIKELIHOOD for discrete part

      for (i in 1:N_discrete){ 
        cov.dis[i] ~ dbern(mu[i])
        logit(mu[i]) <- b_plotgroup[plotgroup.dis[i]] + # ~= random effect of plot group
                        b.compet * compet.dis[i] + 
                        b.shrub_cov * shrub_cov.dis[i] + 
                        b.gramin_cov * gramin_cov.dis[i] + 
                        b.tempXtcws * tempjja.dis[i] * tcws.dis[i] +       # for interaction
                        # b.tempXtcws2 * (tempjja.dis[i]^2) * tcws.dis[i] +  # for interaction
                        b.tcws * tcws.dis[i] + 
                        b.sri * sri.dis[i] +
                        b.tri * tri.dis[i]
      }
      
      
    # LIKELIHOOD for continuous part

      for (j in 1:N_cont){
        cov.cont[j] ~ dbeta(p[j], q[j])
        p[j] <- mu2[j] * phi
        q[j] <- (1 - mu2[j]) * phi
        logit(mu2[j]) <- b_plotgroup[plotgroup.cont[j]] + # ~= random effect of plot group
                        b.compet * compet.cont[j] +
                        b.shrub_cov * shrub_cov.cont[j] +
                        b.gramin_cov * gramin_cov.cont[j] +
                        b.tempXtcws * tempjja.cont[j] * tcws.cont[j] +       # for interaction
                        # b.tempXtcws2 * (tempjja.cont[j]^2) * tcws.cont[j] +  # for interaction
                        b.tcws * tcws.cont[j] + 
                        b.sri * sri.cont[j] +
                        b.tri * tri.cont[j]
      }

      for (k in 1:N_plotgroups){ # length of total plotgroups
        b_plotgroup[k] ~ dnorm(mu.plotgroup[k],tau.plotgroup)
        mu.plotgroup[k] <- intercept + 
                    
                    # plot group level predictors, linear and quadratic term
                    b.tempjja.x * tempjja.tot[k] + 
                    # b.tempjja.x2 * (tempjja.tot[k]^2) + 
                    b.tempcont.x * tempcont.tot[k] + 
                    # b.tempcont.x2 * (tempcont.tot[k]^2) +
                    b.precipjja.x * precipjja.tot[k] # + 
                    # b.precipjja.x2 * (precipjja.tot[k]^2) 
      }
      
      
      # add predicted values (derived parameters)
      for (m in 1:Nxhat){
        phat_compet[m] <- intercept + b.compet * xhat_compet[m]
        phat_graminoid_cover[m] <- intercept + b.gramin_cov * xhat_graminoid_cover[m]
        phat_shrub_cover[m] <- intercept + b.shrub_cov * xhat_shrub_cover[m]
        phat_sri[m] <- intercept + b.sri * xhat_sri[m]
        phat_tri[m] <- intercept + b.tri * xhat_tri[m]
        phat_tcws[m] <- intercept + b.tcws * xhat_tcws[m]
        phat_tempjja[m] <- intercept + b.tempjja.x * xhat_tempjja[m] # + b.tempjja.x2 * (xhat_tempjja[m]^2)
        phat_tempcont[m] <- intercept + b.tempcont.x * xhat_tempcont[m] # + b.tempcont.x2 * (xhat_tempcont[m]^2)
        phat_precipjja[m] <- intercept + b.precipjja.x * xhat_precipjja[m] # + b.precipjja.x2 * (xhat_precipjja[m]^2)
      
        for (p in 1:Nxhat2){
          phat_tempXmoist[m,p] <- intercept +
                                      b.tempjja.x * xhat_tempjja[m] +
                                      # b.tempjja.x2 * (xhat_tempjja[m]^2) +
                                      b.tcws * xhat_tcws2[p] +
                                      b.tempXtcws * xhat_tempjja[m] * xhat_tcws2[p] # +
                                      # b.tempXtcws2 * (xhat_tempjja[m]^2) * xhat_tcws2[p]
          }
        }

    
      }
  ", file.path("..", "models_tcws", "shrub_gradient.SalGla2.jags"))

# specify new set of parameters to be monitored
params_SalGla2 <- c("intercept",
                    "b.tempjja.x", # "b.tempjja.x2",
                    "b.tempcont.x", # "b.tempcont.x2",
                    "b.precipjja.x", # "b.precipjja.x2",
                    "b.compet", 
                    "b.shrub_cov",
                    "b.gramin_cov",
                    "b.sri",
                    "b.tri",
                    "b.tcws",
                    "b_plotgroup[1]","b_plotgroup[2]","b_plotgroup[3]","b_plotgroup[63]",
                    "sigma.plotgroup",
                    "phi",
                    "phat_compet", "phat_shrub_cover", "phat_graminoid_cover", 
                    "phat_sri", "phat_tri", "phat_tcws", 
                    "phat_tempjja", "phat_tempcont", "phat_precipjja",
                    "phat_tempXmoist")

model_out.shrub_gradient.SalGla2 <- jags(shrub_gradient_jags.SalGla.data,   # input data
                                        inits = NULL,                       # JAGS to create initial values
                                        params_SalGla2,                     # parameters to be saved
                                        model.file = file.path("..", "models_tcws", "shrub_gradient.SalGla2.jags"), 
                                        n.chains = 3,                       # no. Markov chains
                                        n.iter = 100000, n.burnin = 70000,  # no. iterations & burn-in fraction per chain
                                        n.thin = 2,                         # thinning rate
                                        DIC = FALSE,                        # do not compute deviance, pD, and DIC
                                        working.directory = NULL, 
                                        progress.bar = "text")

# plot(model_out.shrub_gradient.SalGla2) #check convergence, etc.

Extract coefficients and plot effect sizes:

# extract coefficients 
coeff.shrub_gradient.SalGla2 <- model_out.shrub_gradient.SalGla2$BUGSoutput$summary %>% 
  as.data.frame %>% 
  select('mean','sd','2.5%','97.5%','Rhat') %>% 
# add identifying info to data frame
  rownames_to_column(var = "param")

# add 90% CIs
ci_90.SalGla2 <- data.frame(q5 = NA, q95 = NA, param = NA)
for (param in 1:(length(model_out.shrub_gradient.SalGla$BUGSoutput$sims.list)-4)){
  ci_90.SalGla2[param,1:2] <- quantile(data.frame(model_out.shrub_gradient.SalGla2$BUGSoutput$sims.list[param])[,1], probs = c(0.05, 0.95))
  ci_90.SalGla2[param, 3] <- names(data.frame(model_out.shrub_gradient.SalGla2$BUGSoutput$sims.list))[param]
}

# join to coefficients table
coeff.shrub_gradient.SalGla2 <- coeff.shrub_gradient.SalGla2 %>% 
  left_join(ci_90.SalGla2, by = "param") %>% 
  # reorder and rename cols
  select(param, mean, sd, 
         l95 = "2.5%",
         l90 = q5,
         u90 = q95,
         u95 = "97.5%",
         Rhat) %>% print

save(coeff.shrub_gradient.SalGla2, file = file.path("..", "data", "processed", "model_outputs", "species_tcws", "model_output_SalGla2.Rdata"))
# load(file.path("..", "data", "processed", "model_outputs", "species_tcws", "model_output_SalGla2.Rdata"))

# effect size plot
(effect_size_plot.SalGla2 <- model_plot_sig_function(coeff.shrub_gradient.SalGla2, title_string = "Salix glauca", plot_width = 9.5))

• positive relationship with conservative communities, temperature variability (linear) (sig.)
• slightly positive relationship with other shrub cover and TCwetness (n.s.)

Vaccinium uliginosum

model_out.shrub_gradient.VacUli <- jags(shrub_gradient_jags.VacUli.data,    # input data
                                        inits = NULL,                       # JAGS to create initial values
                                        params,                             # parameters to be saved
                                        model.file = file.path("..", "models_tcws", "shrub_gradient.spec.jags"), 
                                        n.chains = 3,                       # no. Markov chains
                                        n.iter = 100000, n.burnin = 70000,  # no. iterations & burn-in fraction per chain
                                        n.thin = 2,                         # thinning rate
                                        DIC = FALSE,                        # do not compute deviance, pD, and DIC
                                        working.directory = NULL, 
                                        progress.bar = "text")

# plot(model_out.shrub_gradient.VacUli) #check convergence, etc.

Extract coefficients and plot effect sizes:

# extract coefficients 
coeff.shrub_gradient.VacUli <- model_out.shrub_gradient.VacUli$BUGSoutput$summary %>% 
  as.data.frame %>% 
  select('mean','sd','2.5%','97.5%','Rhat') %>% 
# add identifying info to data frame
  rownames_to_column(var = "param")
  # mutate(param = as.vector(sapply(strsplit(rownames(coeff.shrub_gradient.VacUli),"[[]",fixed=FALSE), "[", 1))) #%>% print

# add 90% CIs
ci_90.VacUli <- data.frame(q5 = NA, q95 = NA, param = NA)
for (param in 1:(length(model_out.shrub_gradient.VacUli$BUGSoutput$sims.list)-4)){
  ci_90.VacUli[param,1:2] <- quantile(data.frame(model_out.shrub_gradient.VacUli$BUGSoutput$sims.list[param])[,1], probs = c(0.05, 0.95))
  ci_90.VacUli[param, 3] <- names(data.frame(model_out.shrub_gradient.VacUli$BUGSoutput$sims.list))[param]
}

# join to coefficients table
coeff.shrub_gradient.VacUli <- coeff.shrub_gradient.VacUli %>% 
  left_join(ci_90.VacUli, by = "param") %>% 
  # reorder and rename cols
  select(param, mean, sd, 
         l95 = "2.5%",
         l90 = q5,
         u90 = q95,
         u95 = "97.5%",
         Rhat) %>% print

save(coeff.shrub_gradient.VacUli, file = file.path("..", "data", "processed", "model_outputs", "species_tcws", "first_runs_converged", "model_output_VacUli.Rdata"))
# load(file.path("..", "data", "processed", "model_outputs", "species_tcws", "first_runs_converged", "model_output_VacUli.Rdata"))

# effect size plot
(effect_size_plot.VacUli <- model_plot_sig_function(coeff.shrub_gradient.VacUli, title_string = "Vaccinium uliginosum", plot_width = 12.5))

• positive relationship with more conservative communities and TC wetness (sig.)
• slightly positive relationship with graminoid cover, slightly negative relationship with terrain ruggedness (n.s.)
  

As the quadratic terms for summer temperature, precipitation and temperature variability were not significant, they were removed from the model before re-running:

# new model object with terms removed
write("
  
  model{

    # priors
      
      intercept ~ dnorm(0, 0.0001)
      
      b.compet ~ dnorm(0, 0.0001)
      b.shrub_cov ~ dnorm(0, 0.0001)
      b.gramin_cov ~ dnorm(0, 0.0001)
      b.sri ~ dnorm(0, 0.0001)
      b.tri ~ dnorm(0, 0.0001)
      b.tcws ~ dnorm(0, 0.0001)

      sigma.plotgroup ~ dunif(0,100)
      tau.plotgroup <- 1/(sigma.plotgroup * sigma.plotgroup)
      
      b.tempjja.x ~ dnorm(0, 0.001)
      # b.tempjja.x2 ~ dnorm(0, 0.001)
      b.tempcont.x ~ dnorm(0, 0.001)
      # b.tempcont.x2 ~ dnorm(0, 0.001)
      b.precipjja.x ~ dnorm(0, 0.001)
      # b.precipjja.x2 ~ dnorm(0, 0.001)
      
      b.tempXtcws ~ dnorm(0, 0.001)
      # b.tempXtcws2 ~ dnorm(0, 0.001)
      
      phi ~ dgamma(0.1, 0.1)
      
      
    # LIKELIHOOD for discrete part

      for (i in 1:N_discrete){ 
        cov.dis[i] ~ dbern(mu[i])
        logit(mu[i]) <- b_plotgroup[plotgroup.dis[i]] + # ~= random effect of plot group
                        b.compet * compet.dis[i] + 
                        b.shrub_cov * shrub_cov.dis[i] + 
                        b.gramin_cov * gramin_cov.dis[i] + 
                        b.tempXtcws * tempjja.dis[i] * tcws.dis[i] +       # for interaction
                        # b.tempXtcws2 * (tempjja.dis[i]^2) * tcws.dis[i] +  # for interaction
                        b.tcws * tcws.dis[i] + 
                        b.sri * sri.dis[i] +
                        b.tri * tri.dis[i]
      }
      
      
    # LIKELIHOOD for continuous part

      for (j in 1:N_cont){
        cov.cont[j] ~ dbeta(p[j], q[j])
        p[j] <- mu2[j] * phi
        q[j] <- (1 - mu2[j]) * phi
        logit(mu2[j]) <- b_plotgroup[plotgroup.cont[j]] + # ~= random effect of plot group
                        b.compet * compet.cont[j] +
                        b.shrub_cov * shrub_cov.cont[j] +
                        b.gramin_cov * gramin_cov.cont[j] +
                        b.tempXtcws * tempjja.cont[j] * tcws.cont[j] +       # for interaction
                        # b.tempXtcws2 * (tempjja.cont[j]^2) * tcws.cont[j] +  # for interaction
                        b.tcws * tcws.cont[j] + 
                        b.sri * sri.cont[j] +
                        b.tri * tri.cont[j]
      }

      for (k in 1:N_plotgroups){ # length of total plotgroups
        b_plotgroup[k] ~ dnorm(mu.plotgroup[k],tau.plotgroup)
        mu.plotgroup[k] <- intercept + 
                    
                    # plot group level predictors, linear and quadratic term
                    b.tempjja.x * tempjja.tot[k] + 
                    # b.tempjja.x2 * (tempjja.tot[k]^2) + 
                    b.tempcont.x * tempcont.tot[k] + 
                    # b.tempcont.x2 * (tempcont.tot[k]^2) +
                    b.precipjja.x * precipjja.tot[k] # + 
                    # b.precipjja.x2 * (precipjja.tot[k]^2) 
      }
      
      
      # add predicted values (derived parameters)
      for (m in 1:Nxhat){
        phat_compet[m] <- intercept + b.compet * xhat_compet[m]
        phat_graminoid_cover[m] <- intercept + b.gramin_cov * xhat_graminoid_cover[m]
        phat_shrub_cover[m] <- intercept + b.shrub_cov * xhat_shrub_cover[m]
        phat_sri[m] <- intercept + b.sri * xhat_sri[m]
        phat_tri[m] <- intercept + b.tri * xhat_tri[m]
        phat_tcws[m] <- intercept + b.tcws * xhat_tcws[m]
        phat_tempjja[m] <- intercept + b.tempjja.x * xhat_tempjja[m] # + b.tempjja.x2 * (xhat_tempjja[m]^2)
        phat_tempcont[m] <- intercept + b.tempcont.x * xhat_tempcont[m] # + b.tempcont.x2 * (xhat_tempcont[m]^2)
        phat_precipjja[m] <- intercept + b.precipjja.x * xhat_precipjja[m] # + b.precipjja.x2 * (xhat_precipjja[m]^2)
      
        for (p in 1:Nxhat2){
          phat_tempXmoist[m,p] <- intercept +
                                      b.tempjja.x * xhat_tempjja[m] +
                                      # b.tempjja.x2 * (xhat_tempjja[m]^2) +
                                      b.tcws * xhat_tcws2[p] +
                                      b.tempXtcws * xhat_tempjja[m] * xhat_tcws2[p] # +
                                      # b.tempXtcws2 * (xhat_tempjja[m]^2) * xhat_tcws2[p]
          }
        }

    
      }
  ", file.path("..", "models_tcws", "shrub_gradient.VacUli2.jags"))

# specify new set of parameters to be monitored
params_VacUli2 <- c("intercept",
                    "b.tempjja.x", # "b.tempjja.x2",
                    "b.tempcont.x", # "b.tempcont.x2",
                    "b.precipjja.x", # "b.precipjja.x2",
                    "b.compet",
                    "b.shrub_cov",
                    "b.gramin_cov",
                    "b.sri",
                    "b.tri",
                    "b.tcws",
                    "b_plotgroup[1]","b_plotgroup[2]","b_plotgroup[3]","b_plotgroup[63]",
                    "sigma.plotgroup",
                    "phi",
                    "phat_compet", "phat_shrub_cover", "phat_graminoid_cover", 
                    "phat_sri", "phat_tri", "phat_tcws", 
                    "phat_tempjja", "phat_tempcont", "phat_precipjja",
                    "phat_tempXmoist")

model_out.shrub_gradient.VacUli2 <- jags(shrub_gradient_jags.VacUli.data,   # input data
                                        inits = NULL,                       # JAGS to create initial values
                                        params_VacUli2,                     # parameters to be saved
                                        model.file = file.path("..", "models_tcws", "shrub_gradient.VacUli2.jags"), 
                                        n.chains = 3,                       # no. Markov chains
                                        n.iter = 100000, n.burnin = 70000,  # no. iterations & burn-in fraction per chain
                                        n.thin = 2,                         # thinning rate
                                        DIC = FALSE,                        # do not compute deviance, pD, and DIC
                                        working.directory = NULL, 
                                        progress.bar = "text")

# plot(model_out.shrub_gradient.VacUli2) #check convergence, etc.

Extract coefficients and plot effect sizes:

# extract coefficients 
coeff.shrub_gradient.VacUli2 <- model_out.shrub_gradient.VacUli2$BUGSoutput$summary %>% 
  as.data.frame %>% 
  select('mean','sd','2.5%','97.5%','Rhat') %>% 
# add identifying info to data frame
  rownames_to_column(var = "param")

# add 90% CIs
ci_90.VacUli2 <- data.frame(q5 = NA, q95 = NA, param = NA)
for (param in 1:(length(model_out.shrub_gradient.VacUli2$BUGSoutput$sims.list)-4)){
  ci_90.VacUli2[param,1:2] <- quantile(data.frame(model_out.shrub_gradient.VacUli2$BUGSoutput$sims.list[param])[,1], probs = c(0.05, 0.95))
  ci_90.VacUli2[param, 3] <- names(data.frame(model_out.shrub_gradient.VacUli2$BUGSoutput$sims.list))[param]
}

# join to coefficients table
coeff.shrub_gradient.VacUli2 <- coeff.shrub_gradient.VacUli2 %>% 
  left_join(ci_90.VacUli2, by = "param") %>% 
  # reorder and rename cols
  select(param, mean, sd, 
         l95 = "2.5%",
         l90 = q5,
         u90 = q95,
         u95 = "97.5%",
         Rhat) %>% print

save(coeff.shrub_gradient.VacUli2, file = file.path("..", "data", "processed", "model_outputs", "species_tcws", "model_output_VacUli2.Rdata"))

# effect size plot
(effect_size_plot.VacUli2 <- model_plot_marg_function(coeff.shrub_gradient.VacUli2, title_string = "Vaccinium uliginosum", plot_width = 9.5))

• positive relationship with more conservative communities, TCwetness (sig.) and summer temperature (m.s.)
• slightly positive relationship with graminoid cover and precipitation, negative with terrain heterogeneity (n.s.)